At gamescom in Cologne, Europe's largest games fair, a play­er told me about her expe­ri­ences in ARC Raiders. The game rests on a sim­ple arrange­ment: any­one who gets close enough to you in the world can talk to you. The mechan­ic is called prox­im­i­ty voice, speak­ing with­in earshot.

An encounter can turn into almost any­thing: an alliance, a trade, an ambush. Nobody has to talk. But who­ev­er talks can be heard.

She talked. A few words were enough.

Anoth­er play­er, by her account, could tell from her voice that she is a woman. From that moment she was no longer an oppo­nent like any oth­er, but a tar­get. He fol­lowed her across the entire map for the rest of the round, try­ing to kill her. Nobody has to guess at the motive. He told her: because she is a woman.

The obvi­ous read­ing is a mod­er­a­tion fail­ure. Some­where a fil­ter was miss­ing, a work­ing report but­ton, a sanc­tion, and once those are added, the prob­lem is solved.

Beside it stands a sec­ond read­ing, which does not replace the first. A sys­tem whose appeal comes from unmod­er­at­ed, anony­mous encoun­ters pro­duces rounds like this one not only as a dis­tur­bance but as a run­ning cost. The ten­sion is inten­tion­al; the stu­dio says so open­ly. The ques­tion is who pays.

Her case is one case. The pat­tern behind it is well doc­u­ment­ed. As ear­ly as 2013, Kuznekoff and Rose showed in a field exper­i­ment involv­ing over 1,600 fel­low play­ers that a female voice in voice chat drew, on aver­age, three times as many neg­a­tive com­ments as a male voice or no voice at all [1].

So the ques­tion is not whether a voice becomes a tar­get. The ques­tion is what stands between the voice and the attack. The answer varies from game to game, but it comes down to four modes, one word each: mute, report, record, lis­ten in.

Mute: Silence It Yourself

Mut­ing is in the play­er's own hands. The tar­get­ed play­er removes the attack­er from her own head­phones. Every­one else on the serv­er can still hear him, and for her he is still in the game. No sanc­tion fol­lows, and noth­ing is doc­u­ment­ed.

Gray Zone War­fare serves as the exam­ple for this mode. When audio pro­gram­mer Roman Dufek pre­sent­ed the game's voice sys­tem at dev­com 2026, the devel­op­er con­fer­ence that runs along­side gamescom, exact­ly one lay­er came up on the ques­tion of voice mod­er­a­tion: play­er-side mut­ing [2]. Report­ing chan­nels exist; more on that short­ly. In the moment itself, only mut­ing has any effect.

The archi­tec­ture makes it worse. Any­one talk­ing to their own team over the in-game radio is also trans­mit­ting on prox­im­i­ty voice; there is no tru­ly pri­vate chan­nel in the field. You can­not step out of the open room. You can only turn it down, one voice at a time.

Rust is no excep­tion. The game has shaped its genre for over a decade and is noto­ri­ous for its tone, and on its own servers it goes a step fur­ther. Facepunch, the stu­dio behind it, writes that hate and slurs against whole groups are con­tin­u­ous­ly mon­i­tored there and auto­mat­i­cal­ly flagged to its staff, who then review and decide. Bans are per­ma­nent and are not lift­ed [26]. Whether that mon­i­tor­ing cov­ers voice chat is stat­ed nowhere. The page does not dis­tin­guish text from speech.

The page also says this: a banned play­er can keep play­ing on com­mu­ni­ty servers. And on those, which is where most of the game hap­pens, the rules belong to who­ev­er runs the serv­er; how dif­fer­ent­ly those spaces are run is some­thing I have writ­ten about else­where [8].

For the play­er in the moment, the tool is the same. The sup­port page on the sub­ject is titled "Mut­ing and Cen­sor­ship". It opens by say­ing that, as in most online games, you will meet your share of less agree­able play­ers; what it offers is mut­ing, which cuts text and voice at once [26]. The first answer to the harassed play­er is the same one: silence it your­self.

Report: The Button and What Comes After

You can report in all three games: ARC Raiders, Gray Zone War­fare, Rust. Rust has a key that opens a form; Gray Zone War­fare has had an in-game sys­tem since Novem­ber 2024, plus a sup­port tick­et for which the stu­dio explic­it­ly asks for evi­dence [24][26]. But the mode does not begin at the but­ton. It begins with what hap­pens next.

On that, the three stu­dios say dif­fer­ent things.

Embark, the stu­dio behind ARC Raiders, states that every report is reviewed and ver­i­fied by its team before any action is tak­en, and that mass report­ing alone gets nobody banned; that address­es a known form of abuse, the coor­di­nat­ed report­ing of indi­vid­ual play­ers [3].

Facepunch says the same about man­u­al review on its own servers [26].

MADFINGER, the stu­dio behind Gray Zone War­fare, writes in its terms of use that hate speech, harass­ment and threats lead to tem­po­rary or per­ma­nent bans, encour­ages play­ers to report, and in the same para­graph makes clear that false report­ing can be pun­ished too [24]. How reports are reviewed, it does not say.

Embark and MADFINGER address abuse at oppo­site ends: Embark on the side of the report­ed, MADFINGER on the side of the reporter.

And so the ques­tion that car­ries the whole mode stays open: what is pre­served from a spo­ken inci­dent, so that a report can be checked at all? On that, all three stu­dios are silent.

Record: The Studio Keeps the Evidence

For Val­o­rant, Riot Games took the step the three stu­dios above have not: the game records voice chat. All record­ings are kept for 24 hours. If no report comes in, they are delet­ed. If one does, the reten­tion peri­od is sev­en days; if a penal­ty fol­lows, it can rise in Europe, depend­ing on the penal­ty's length, to two years [4]. Secur­ing the spo­ken evi­dence is now the stu­dio's job. The peo­ple affect­ed do not have to record it them­selves.

What is remark­able is how slow­ly this arrived. The pri­va­cy notice was amend­ed in 2021; from July 2022 record­ing ran in North Amer­i­ca in the back­ground; in Feb­ru­ary 2023 the Eng­lish-lan­guage beta start­ed there. For Europe, the same beta was not announced until Sep­tem­ber 2024, a good year and a half lat­er [5].

A sec­ond stu­dio has kept the same path open with­out describ­ing it. Embark's pri­va­cy pol­i­cy lists com­mu­ni­ca­tion between play­ers dur­ing a ses­sion, express­ly includ­ing voice, as a cat­e­go­ry of data processed to pur­sue pro­hib­it­ed activ­i­ties. Both man­u­al and auto­mat­ed pro­ce­dures are used, Embark says, with sanc­tions up to a per­ma­nent ban [3]. Whether and what ARC Raiders actu­al­ly records, the pol­i­cy does not say. Riot tells you what it records and for how long. Embark tells you what pro­cess­ing it intends.

Listen In: The Machine on the Channel

Since 2023, Call of Duty has had a machine lis­ten­ing. Tox­Mod, built by the ven­dor Mod­u­late, eval­u­ates voice chat con­tin­u­ous­ly, with­out wait­ing for a report, and by now does so in Europe and in Ger­man as well [6][7].

The fig­ures come from Activi­sion and Mod­u­late them­selves and have not been inde­pen­dent­ly checked, but they are clear: expo­sure to tox­ic speech fell con­sis­tent­ly by 25 to 33 per­cent in tests, and repeat-offence rates among sanc­tioned play­ers fell, by the pub­lish­er's account, by 43 and 38 per­cent with­in a year, depend­ing on the title [7]. How the two sides talk about the sys­tem them­selves, Mark Frumkin of Mod­u­late and Grant Cahill of Activi­sion, I wrote up after their joint talk at dev­com 2025 [8].

That mod­er­a­tion works in this game is by now also an aca­d­e­m­ic find­ing. A study with authors from Cal­tech and Activi­sion found for Mod­ern War­fare II that sanc­tions reduce dis­rup­tive behav­iour and, at the same time, dri­ve some of the offend­ing play­ers out of the game alto­geth­er [30]. It assess­es mod­er­a­tion as a whole, not one prod­uct.

One more fig­ure from the ven­dor's case study shows the lim­its of play­er reports. For rough­ly half of them, no audio of the report­ed play­er exist­ed. Of the reports that did have audio, an esti­mat­ed half con­tained a vio­la­tion. In the end, only about a quar­ter of reports came with audio that doc­u­ment­ed a vio­la­tion [7]. What lies behind the uncon­firmed reports, the case study does not break down.

Four Modes, Not a Ladder

You could read the four as a time­line, from the mute but­ton to AI mod­er­a­tion. That is not how they work. They stack: a game that lis­tens in also lets you mute and report. And they are not stages of devel­op­ment but deci­sions that sit side by side on the mar­ket, in the same genre, some­times on the same tech­nol­o­gy.

"Can't Be Done" Comes Off the Shelf

So why does a game stop at mut­ing and report­ing? Two expla­na­tions sug­gest them­selves: there are no effec­tive tools, or there are tools only for the big play­ers. Both can be checked.

What Developers Say Themselves

The first expla­na­tion can be test­ed against devel­op­ers. In a sur­vey com­mis­sioned by Uni­ty among some 400 devel­op­ers in the Unit­ed States, the Unit­ed King­dom and South Korea, 96 per­cent agreed that effec­tive tools against tox­ic behav­iour exist. Which of those tools specif­i­cal­ly cov­er voice chat, the fig­ures do not say. Only 52 per­cent named invest­ment in such tech­nol­o­gy a pri­or­i­ty for their stu­dio that year, and 56 per­cent admit­ted that pro­tect­ing play­ers from tox­ic behav­iour is not at the top of their list. What has been imple­ment­ed is most­ly the obvi­ous: report­ing func­tions at 54 per­cent, bans at 47 per­cent [28].

What the Supply Chain Offers

That leaves the sec­ond expla­na­tion, and it does at least describe the sit­u­a­tion of a few years ago: proac­tive voice mod­er­a­tion was enter­prise tech­nol­o­gy, built for the largest pub­lish­ers. That is chang­ing now, in the place where small stu­dios shop: the infra­struc­ture.

Since March 2025, Vivox has had the mod­er­a­tion ven­dor GGWP on board as its safe­ty part­ner [9]. Vivox is Uni­ty's voice-chat com­po­nent, with serv­er-side inte­gra­tion for every Vivox cus­tomer, and it sits inside a large num­ber of inde­pen­dent and mid-sized titles. The first promi­nent deploy­ment of the voice mod­er­a­tion is Goril­la Tag, a VR game from an inde­pen­dent stu­dio [9]. The entry bar­ri­er drops below the AAA seg­ment. What an inte­gra­tion costs, how well it detects in which lan­guage, and who reviews the hits is anoth­er mat­ter; mod­er­a­tion does not become cheap or casu­al because of it.

Clos­er to one of the three games above: 4Players, the ven­dor behind ODIN, the voice tech­nol­o­gy in Gray Zone War­fare, now sells the capa­bil­i­ty itself. Its serv­er-side SDK explic­it­ly adver­tis­es audio access for mod­er­a­tion tools [10]. With Cor­tex, 4Players offers a prod­uct of its own: real-time tran­scrip­tion of every voice stream, auto­mat­ic detec­tion of hate speech and harass­ment, sanc­tions pushed back into the room, cur­rent­ly in ear­ly access [12]. Gray Zone War­fare thus runs on a plat­form whose mak­er has lis­ten­ing-in in its cat­a­logue, if only as a prod­uct that is not fin­ished itself.

One caveat belongs here. ODIN offers option­al end-to-end encryp­tion. With­out access to the keys, the relay­ing serv­er can­not eval­u­ate the speech [10]. How Gray Zone War­fare routes its audio, the stu­dio sketched at dev­com: ODIN deliv­ers the voice data, an adapter pass­es it to the audio engine Wwise, and the code for that is pub­lic on GitHub [2][11]. Whether auto­mat­ed eval­u­a­tion could hook into that path was left open on stage.

What Would Be Possible in ARC Raiders

ARC Raiders' voice chat runs on Epic Online Ser­vices [19], and on exact­ly this infra­struc­ture Epic has built voice report­ing for its own Fort­nite. With voice report­ing enabled, par­tic­i­pants' devices buffer rough­ly the last five min­utes of dig­i­tal­ly signed voice data. There is no con­tin­u­ous serv­er-side record­ing. Only a report trans­mits the clip to Epic; sep­a­rate reten­tion peri­ods then apply for review and sanc­tions [25]. This kind of evi­dence cap­ture, a local buffer and an upload only when some­thing hap­pens, is also worth not­ing for how lit­tle it stores.

The sys­tem is not doc­u­ment­ed as a ready-made fea­ture for oth­er Epic cus­tomers, but the audio access it rests on is [25]. The tech­ni­cal prin­ci­ple is pub­lic, and it runs on the same infra­struc­ture ARC Raiders uses. Whether ARC Raiders uses any­thing com­pa­ra­ble, Embark does not say. Its pri­va­cy pol­i­cy does pro­vide for pro­cess­ing such com­mu­ni­ca­tion.

Seen from out­side, this much can be said: any­one who stops at mut­ing and report­ing today can no longer claim there is noth­ing else. Which pro­ce­dures these stu­dios actu­al­ly use, and why, is stat­ed nowhere in pub­lic. It is exact­ly the kind of ques­tion one can put to them.

The Evidence That Must Not Exist

Reporting Without Evidence

Back to report­ing. There is a trap here, and you only see it once you play it through.

A play­er is harassed on prox­im­i­ty voice. She reports it, and what is labo­ri­ous about that even in the ordi­nary case I have described else­where: open the menu, fill in the form, in the mid­dle of a run­ning match [8]. With speech, some­thing else comes on top.

A hate­ful com­ment stays where it is. You can link to it, screen­shot it, show it to the review­er. The pro­ce­dure for report­ing ille­gal con­tent to host­ing ser­vices under the EU's Dig­i­tal Ser­vices Act fol­lows the same log­ic: Arti­cle 16 requires an expla­na­tion of why the con­tent is con­sid­ered ille­gal and a pre­cise indi­ca­tion of its elec­tron­ic loca­tion, such as a URL [13].

A sen­tence that fell into your head­phones twen­ty sec­onds ago has no loca­tion. It is gone.

So review­ers at the stu­dio have lit­tle in hand when the game does not record. One word against anoth­er. MADFINGER states the expec­ta­tion open­ly and asks explic­it­ly for video record­ings. In the same forum thread, a mod­er­a­tor explains how to switch on Steam's back­ground record­ing [24]. The tech­ni­cal answer is there. Whether such a record­ing is law­ful in a giv­en case, the instruc­tions do not say.

What the Player May Do

So next time she records in the back­ground her­self. And runs into the next sur­prise: it might be a crim­i­nal offence.

The play­er I spoke to plays under Ger­man law, and this sec­tion fol­lows that law. Any­one in Ger­many who records anoth­er per­son­'s non-pub­lic spo­ken word with­out autho­ri­sa­tion, or pass­es such a record­ing on, faces under Sec­tion 201 of the Crim­i­nal Code up to three years in prison or a fine [14]. The wish to secure evi­dence does not auto­mat­i­cal­ly help. In a civ­il case, Ger­many's Fed­er­al Court of Jus­tice held that a gen­er­al inter­est in obtain­ing evi­dence does not by itself jus­ti­fy the intru­sion into anoth­er per­son­'s per­son­al­i­ty rights [16]. The crim­i­nal ques­tion stands beside that, untouched. Oth­er juris­dic­tions draw the line dif­fer­ent­ly; in the Unit­ed States, for instance, whether a record­ing is law­ful depends on the state and on whose con­sent is required. The pat­tern holds every­where: the per­son who records to defend her­self is the one car­ry­ing the legal risk.

The rule pro­tects some­thing sen­si­ble: what I say when the pub­lic is not lis­ten­ing should not be stored and kept with­out my knowl­edge. For a con­ver­sa­tion in a liv­ing room or on the phone, that pro­tec­tion is imme­di­ate­ly intu­itive.

But what about the voice chat of a video game? Only those who come close enough can hear the spo­ken word. The game serv­er itself, though, is open to any­one. Does the pro­tec­tion apply here or not? In the sources I searched, I have found no court deci­sion on whether prox­im­i­ty voice counts as non-pub­lic speech in this sense. Nor any legal pub­li­ca­tion that address­es this spe­cif­ic ques­tion [15].

Any­one who wants to defend her­self with her own record­ing car­ries the legal risk alone. With­out a record­ing, what remains in a dis­pute are state­ments: her own, and, if some­one hap­pened to be lis­ten­ing, those of third par­ties.

What the Studio May Do

More than most peo­ple think. Intu­ition says: any­one who records my voice has to ask me first. But Euro­pean data pro­tec­tion law knows a sec­ond route besides con­sent. A com­pa­ny may process data if it has a legit­i­mate inter­est of its own, if the pro­cess­ing is nec­es­sary for that inter­est, and if that inter­est out­weighs the rights of the peo­ple con­cerned. That is exact­ly what Riot relies on in Europe, cit­ing its inter­est in a safe gam­ing envi­ron­ment [27].

Riot thus bases pro­cess­ing for safe­ty pur­pos­es on legit­i­mate inter­est, not on con­sent. Any­one who wants to avoid hav­ing their voice com­mu­ni­ca­tion cap­tured must, by Riot's account, dis­able voice chat [4][27].

Whether that bal­anc­ing test legal­ly sup­ports con­tin­u­ous record­ing of every con­ver­sa­tion is not set­tled. It is Riot's posi­tion, not a reg­u­la­tor's con­fir­ma­tion, and it does not answer the sep­a­rate ques­tion of crim­i­nal law. Riot itself words this care­ful­ly and reserves the right to lis­ten only "where legal­ly per­mis­si­ble" [27].

Whether these open ques­tions explain why Riot took a year and a half longer in Europe than in North Amer­i­ca is not doc­u­ment­ed. Riot attrib­ut­es the staged roll­out to the matu­ri­ty of its lan­guage mod­els; that does not explain Europe's lag [5].

Who Is Willing to Be Recorded

Some­thing else belongs here before record­ing is pre­sent­ed as the clean solu­tion. In the Uni­ty sur­vey men­tioned above, 68 per­cent of play­ers said they were will­ing to be record­ed if it helps against tox­i­c­i­ty. At the same time, 89 per­cent expressed con­cerns: they want to be sure the data is not mis­used, sold on or leaked [28].

The same peo­ple, then, are will­ing and wor­ried at once. And the will­ing­ness is uneven­ly dis­trib­uted: in the Unit­ed States and the Unit­ed King­dom it is low­er among women than among men [28]. That is the group whose voic­es, as mea­sured, draw the most hos­til­i­ty in games [1].

The result is an asym­me­try. A game that does not record itself avoids one more copy of the con­ver­sa­tion.

That sounds like data pro­tec­tion. But it has a price:

• In prac­tice, that con­fi­den­tial­i­ty also pro­tects the harass­er who would rather have his words doc­u­ment­ed nowhere if a jus­ti­fied report comes in.
• And it bur­dens the peo­ple affect­ed, who would have to pre­serve those words to be tak­en seri­ous­ly. Who­ev­er for­goes record­ing also shifts the legal risk: away from the stu­dio, onto the peo­ple affect­ed.

Disguise as a Feature

A Tool With Two Sides

There is one more way out, and it comes from a stu­dio itself. ARC Raiders ships with a built-in voice chang­er: Raider Voice, a real-time fil­ter on your own micro­phone, ten voic­es to choose from, five read as male and five as female [17]. Any­one who wants to can speak with a voice that is read dif­fer­ent­ly from their own.

It is not just a fun fea­ture. Design direc­tor Vir­gil Watkins describes Raider Voice as an offer for peo­ple who do not feel com­fort­able speak­ing as them­selves in this envi­ron­ment [18]. In the devel­op­er inter­view with Epic, the phrase "safe space" comes up [19]. The stu­dio has under­stood that its core mechan­ic car­ries a risk.

And a voice fil­ter can be more than a shield. For some it is expres­sion, not cam­ou­flage; trans and non-bina­ry users of social VR report that a voice chang­er lets them sound the way they want to sound, and that it reduces voice-relat­ed harass­ment along the way [21].

Does the Offer Become a Standard?

What deserves crit­i­cism is not that the tool exists. What would deserve crit­i­cism is its exis­tence turn­ing into an expec­ta­tion. And you have to look at where it points. Not at the harass­er, but at the harassed. She is the one who is sup­posed to dis­guise her­self so that the game works for her the way it works for every­one else.

There may be a sec­ond, qui­eter shift in this. As long as there is no fil­ter, an audi­ble female voice is just a voice. Once there is one, it can become, in some ears, some­thing that could have been avoid­ed. Noth­ing changes about her; what changes is how oth­ers read her: who­ev­er makes her­self audi­ble as a woman has, by this log­ic, cho­sen to.

There is a term for this pat­tern, and I have used it else­where before: safe­ty work. It describes the every­day safe­ty labour women per­form before any­thing has hap­pened at all: the keys held between the fin­gers to fend off a pos­si­ble attack, the head­phones that are meant to sig­nal that you are not to be approached [33].

The crim­i­nol­o­gists Liz Kel­ly and Fiona Vera-Gray sum up the mes­sage behind it in a sequence that sounds dif­fer­ent in a text about voice chan­nels than in one about parks: women are to be less, in order to be safe. Less vis­i­ble, less free, less vocal [32].

Two of their find­ings belong here. First, this labour turns from an option into a require­ment, not some­thing a woman does but some­thing she is; who­ev­er fails to per­form it is quick­ly seen not as some­one who did some­thing wrong but as some­one who is wrong. Sec­ond: the suc­cess of this labour can­not be shown. When cau­tion works, noth­ing hap­pens, and noth­ing is not evi­dence. What can be count­ed are only the cas­es where cau­tion was not enough.

That is why one can always ask after­wards why she did not sim­ply use the fil­ter. It is the same form of ques­tion that else­where asks what she was wear­ing. A voice fil­ter is safe­ty work in dig­i­tal form. The scale is not the same; the pat­tern is. The game ships the tool. The work of remem­ber­ing it in time stays with the play­er.

Whether that hap­pens in ARC Raiders, nobody has exam­ined as far as I know. It cer­tain­ly does not describe the pur­suer from the begin­ning; he named his motive him­self. It describes a risk that every pro­tec­tive offer car­ries: that an offer becomes a stan­dard against which only one group is mea­sured.

Does It Even Work?

This is where it gets thin. Voice research has known for a long time what it takes to change the per­ceived gen­der of a voice: as a rule, it is not enough to pitch the voice low­er or high­er. Both pitch and tim­bre play a role in how a voice is assigned a gen­der [20]. Whether Raider Voice changes those fea­tures enough for oth­er play­ers to read the voice dif­fer­ent­ly is not answered by that, and not by the stu­dio either.

I find no inde­pen­dent test of whether the fil­ters reli­ably change per­ceived gen­der. That is remark­able, because every­thing around this ques­tion is well researched. That female voic­es in games draw many times the hos­til­i­ty has been mea­sured [1]. That women dis­guise them­selves in games or avoid com­mu­ni­ca­tion alto­geth­er to escape it is wide­ly doc­u­ment­ed, up to and includ­ing reach­ing for a voice chang­er [21]. What is miss­ing is a test of the tools the stu­dios ship for it.

For the play­er from the begin­ning, this means: the pro­tec­tion the game offers her in the moment is one whose effect nobody has demon­strat­ed. And she did not have it on. She played on default set­tings, and the default set­ting is your own voice. The choice one might attribute to her, by the log­ic above, she nev­er made.

You have to know Raider Voice exists, find it and switch it on [17], which means know­ing before your first sen­tence that you are going to need it. That spells out more pre­cise­ly what she is expect­ed to bring: the dis­guise, the knowl­edge that it is nec­es­sary, and the will­ing­ness not to sound like her­self in order to play the same game as every­one who gives up noth­ing for it.

Help Within Earshot

One could raise a sim­ple demand at this point: get rid of the open chan­nel, or at least allow voice only when it is record­ed and machine-mon­i­tored. Before sign­ing that, it is worth look­ing at what else the same chan­nel car­ries.

I used Gray Zone War­fare above as the exam­ple for mut­ing. The same game showed me anoth­er side in ear­ly sum­mer at Game Access 2026, a devel­op­er con­fer­ence in Brno, Czech Repub­lic. At the booth, com­mu­ni­ca­tions lead Jakub Vam­ber­ský showed me the game's field man­u­al; the answer to how you climb a par­tic­u­lar wall was in there, but it took sev­er­al search­es to find [22]. And for any­one the man­u­al can­not help, he named the most impor­tant key in the game: Z, push-to-talk, straight into the server's voice chat, where the com­mu­ni­ty takes new­com­ers by the hand [22].

The open chan­nel that car­ries unmod­er­at­ed harass­ment is the same one over which strangers help each oth­er. It is attack sur­face and sup­port infra­struc­ture at once, and who­ev­er removes it los­es both.

This is no objec­tion to every­thing above. Help­ful­ness toward any­one who is stuck and hos­til­i­ty toward the one who becomes audi­ble as a woman do not exclude each oth­er. Both can live in the same com­mu­ni­ty, even in the same peo­ple.

But it shifts the ques­tion. It is no longer: open chan­nel, yes or no. It is: how much is the open chan­nel worth to a stu­dio, and who pays the price as long as it keeps the chan­nel with­out secur­ing it?

Priced In

What No Mode Delivers

One lim­it con­nects the four modes: none of them guar­an­tees that a sit­u­a­tion like the one at the begin­ning ends imme­di­ate­ly.

For report­ing and record­ing that is obvi­ous; they act after­wards. With lis­ten­ing in, one has to be more pre­cise. Mod­u­late explic­it­ly adver­tis­es Tox­Mod as detect­ing behav­iour while it hap­pens and trig­ger­ing warn­ings or auto­mat­ed actions in real time [34]. The case study describes a three-day shad­ow mute as the sec­ond esca­la­tion step [7]. So it is not out of the ques­tion that a sys­tem recog­nis­es a sen­tence like the one he said to her dur­ing the match and cuts the voice. How far Activi­sion auto­mat­i­cal­ly inter­venes in run­ning rounds, the pub­lic mate­r­i­al does not say.

But even a mute would not have end­ed the pur­suit. It would have tak­en his voice, not his chase. Who­ev­er fol­lows some­one across the map to kill them does it with the means of the game, and no voice mod­er­a­tion sys­tem is respon­si­ble for that. What the modes gov­ern is what was said. What hap­pened to her was more than that.

Yet the game gen­er­ates data on posi­tions and com­bat at the same time. That could be com­bined with voice data, giv­en the inter­faces and a law­ful basis for the pro­cess­ing. Ven­dors such as Get­gud already describe plat­forms that bring voice tran­scripts and game­play behav­iour togeth­er. Pub­licly ver­i­fi­able evi­dence that this reli­ably detects and inter­rupts a tar­get­ed pur­suit like the one described here, I have not found [35]. Patents describe the joint eval­u­a­tion of com­mu­ni­ca­tion and game­play too. Sony has patent­ed mod­er­a­tion sys­tems that com­bine com­mu­ni­ca­tion with what hap­pens in the game [35].

The open ques­tion thus reach­es beyond the four voice modes: how can a spo­ken threat be assessed togeth­er with the behav­iour that fol­lows it in the game? Voice mod­er­a­tion alone offers no com­pre­hen­sive pro­tec­tion against harass­ment in play. That is a rea­son to add the view of in-game behav­iour to it, not a rea­son to drop it.

Who Doesn't Record

Tech­ni­cal options beyond mut­ing and report­ing are avail­able, by now down to the cat­a­logues of mid­dle­ware ven­dors. Whether and how they can be inte­grat­ed and run law­ful­ly depends on the spe­cif­ic game. It is also a deci­sion about how much risk and effort a stu­dio is will­ing to car­ry.

So it is a deci­sion when a game stops, for its voice chat, at mut­ing and report­ing. For ARC Raiders, Gray Zone War­fare and Rust, nowhere is it doc­u­ment­ed that what is said is pre­served or lis­tened to. For Rust, Facepunch explic­it­ly assigns respon­si­bil­i­ty for com­mu­ni­ty servers to their oper­a­tors, includ­ing action against rule vio­la­tions. The stu­dio sets over­ar­ch­ing rules and reserves sanc­tions against servers; uni­form record­ing or eval­u­a­tion of voice chat is not promised [26].

All three draw part of their appeal from the meet­ing of two strangers' voic­es: from the deci­sion whether to trust the oth­er voice or not. What record­ing or review hap­pens in the back­ground stays unclear on the pub­licly avail­able infor­ma­tion.

Three cas­es are not a pat­tern. But they sug­gest a ques­tion.

The Bill

This ten­sion is no acci­dent. It is the prod­uct. Games­Beat, which played two hours with Embark CEO Patrick Söder­lund, sums up the archi­tec­ture of ARC Raiders this way: every­thing is designed to cre­ate ten­sion; you can declare your­self "friend­ly" over voice, and the oth­er side does not have to believe you [23]. Design direc­tor Vir­gil Watkins describes it as a design prin­ci­ple: you do not design AI encoun­ters and play­er con­flicts sep­a­rate­ly, but sit­u­a­tions in which both sys­tems col­lide [31]. Söder­lund calls it the nature of the mul­ti­play­er game: the friend­liest play­er can start shoot­ing at any moment, and that can­not be curat­ed [29].

No stu­dio says it for­goes voice mod­er­a­tion to pre­serve this ten­sion. What can be observed is the com­bi­na­tion: the open chan­nel is part of the prod­uct, and what is pre­served from a spo­ken harass­ment so that a report can be checked, none of the three stu­dios says.

Per­haps it is cost, or staff, or an ear­ly access in which oth­er things are press­ing, or legal cau­tion. From the per­spec­tive of the play­er who was hunt­ed for a round, that changes lit­tle. She car­ried the cost of this arrange­ment, not those who set it up.

That is the bill that appears in no inter­view.

This is the Eng­lish ver­sion of "In Hör­weite, außer Reich­weite". It is a free ren­der­ing, not a mir­ror.

A note on sourc­ing: For this text I did not ask Embark Stu­dios, Facepunch or MADFINGER Games for gen­er­al state­ments on record­ing, evi­dence and inte­gra­tion hur­dles; inquiries on indi­vid­ual points are pend­ing, and I will add answers with a date. Where this text says a stu­dio does not doc­u­ment some­thing, it means exact­ly that: it is not pub­licly doc­u­ment­ed. Three ques­tions remain open and would mer­it a study of their own. First, whether built-in voice fil­ters actu­al­ly change per­ceived gen­der; I have found no pub­lished lis­ten­ing test for Raider Voice. Sec­ond, whether prox­im­i­ty voice legal­ly counts as non-pub­lic speech; in the sources I searched, I found no court deci­sion on this. Third, how reli­ably sys­tems that eval­u­ate voice and game­play data togeth­er can con­nect a spo­ken threat with the pur­suit that fol­lows it. The legal pas­sages fol­low Ger­man law; read­ers else­where will find the rules dif­fer and the pat­tern the same.

Sources (expand here)

[1] Kuznekoff, J. H. / Rose, L. M. (2013): Com­mu­ni­ca­tion in mul­ti­play­er gam­ing: Exam­in­ing play­er respons­es to gen­der cues. New Media & Soci­ety 15(4), 541-556. Halo 3, 245 match­es, 1,660 play­ers, pre-record­ed clips; the female voice drew on aver­age three times as many neg­a­tive com­ments. https://doi.org/10.1177/1461444812458271

[2] Roman Dufek (MADFINGER Games): ses­sion on the voice sys­tem of Gray Zone War­fare, dev­com 2026, Cologne. Author's notes.

[3] Embark Stu­dios, Help Cen­ter: Play­er Report­ing (as of 27 Jan 2026). https://id.embark.games/arc-raiders/support/faq/166-player-reporting. ARC Raiders Pri­va­cy Pol­i­cy (effec­tive 1 June 2026; sup­port page as of 6 July 2026): sup­port page https://id.embark.games/arc-raiders/support/faq/128-privacy-policy, linked pol­i­cy https://m.nexon.com/terms/831. Under "Ana­lyze and address pro­hib­it­ed activ­i­ties", cor­re­spon­dence between play­ers dur­ing a ses­sion includ­ing voice com­mu­ni­ca­tion ("to the extent avail­able") is list­ed as a data cat­e­go­ry, with a note on man­u­al and auto­mat­ed deci­sion-mak­ing up to per­ma­nent bans; legal basis legit­i­mate inter­est under Art. 6(1)(f) GDPR. The same cat­e­go­ry appears under "Oper­ate and pro­vide the Ser­vices" on a con­trac­tu­al basis, Art. 6(1)(b). Reten­tion: only as long as nec­es­sary, for the dura­tion of the rela­tion­ship and at most one year beyond. The doc­u­ment describes per­mis­sions, not prac­tice, and applies to all Embark ser­vices.

[4] Riot Games: VALORANT Voice Eval­u­a­tion Update (June 2022, lat­er amend­ed): reten­tion 24 hours, 7 days on report, up to 2 years on penal­ty. https://playvalorant.com/en-us/news/announcements/valorant-voice-evaluation-update/

[5] Riot time­line: pri­va­cy notice change, April 2021, https://www.riotgames.com/en/news/updating-the-privacy-notice-and-terms-of-service; announce­ment of the NA beta, Feb­ru­ary 2022, https://playvalorant.com/en-us/news/dev/valorant-systems-health-series-voice-and-chat-toxicity/; North Amer­i­ca start 13 July 2022 [4]; APAC and LATAM expan­sion announced in the updat­ed ver­sion of [4]; Eng­lish-only beta for EMEA and Brazil announced in the updat­ed ver­sion of [4], dis­trib­uted via the offi­cial VALORANT chan­nel on 23 Sept 2024 (https://x.com/VALORANT/status/1838247091816271964); "mid-Octo­ber" date giv­en in the same chan­nel's post of 11 Oct 2024 (https://x.com/VALORANT/status/1844769938524786756).

[6] Activi­sion: Call of Duty Voice Chat Mod­er­a­tion FAQ (Tox­Mod by Mod­u­late; voice chat is mon­i­tored and record­ed for mod­er­a­tion; start August 2023 in North Amer­i­ca, glob­al exclud­ing Asia since Mod­ern War­fare III, Novem­ber 2023). https://support.activision.com/articles/call-of-duty-voice-chat-moderation

[7] Modulate/Activision: case study on Tox­Mod in Call of Duty (25 to 33 per­cent less tox­i­c­i­ty expo­sure in com­par­i­son tests; around a quar­ter of play­er reports came with audio doc­u­ment­ing a vio­la­tion), https://www.modulate.ai/case-studies/modulate-activision-case-study, PDF: https://22677265.fs1.hubspotusercontent-na1.net/hubfs/22677265/Collateral/Modulate%20Activision%20Case%20Study.pdf; Activi­sion: Play­er Safe­ty & Pos­i­tive Play, Progress Report Novem­ber 2025 (repeat-offence rates since Decem­ber 2024 down 43 per­cent in Black Ops 6 and 38 per­cent in War­zone; proac­tive voice mod­er­a­tion in Europe, Ger­man since Black Ops 6), https://www.callofduty.com/blog/2025/11/call-of-duty-black-ops-7-black-ops-6-warzone-player-safety-positive-play; Progress Report Octo­ber 2024, https://www.callofduty.com/blog/2024/10/call-of-duty-anti-toxicity-progress-report-black-ops-6-moderation-results. All fig­ures are the pub­lish­er's and ven­dor's own.

[8] Björn Burghard: Tox­ic Behav­ior in Online Games: What Actu­al­ly Helps. Way­cleft, 3 March 2026, last updat­ed 14 June 2026. https://waycleft.com/toxic-behavior-in-online-games-what-actually-helps/ (Ger­man ver­sion: https://waycleft.com/de/toxisches-verhalten-in-online-spielen-was-dagegen-hilft/). Includes the write-up of the joint talk by Mark Frumkin (Mod­u­late) and Grant Cahill (Activi­sion), dev­com 2025, Cologne.

[9] GGWP: GGWP Adds Proac­tive Voice Mod­er­a­tion to be the Safe­ty Part­ner for Uni­ty's Vivox Voice Chat, press release, Busi­ness Wire, 17 March 2025 (serv­er-side inte­gra­tion for all Vivox cus­tomers; Goril­la Tag by Anoth­er Axiom). https://www.businesswire.com/news/home/20250312575564/en/GGWP-Adds-Proactive-Voice-Moderation-to-be-the-Safety-Partner-for-Unitys-Vivox-Voice-Chat. Games­Beat, 17 June 2025 (Vivox pow­ers voice chat for the major­i­ty of mul­ti­play­er games, accord­ing to GGWP): https://venturebeat.com/games/ggwp-adds-voice-moderation-for-unitys-vivox-voice-chat-and-gorilla-tag/

[10] 4Players, ODIN doc­u­men­ta­tion: Node.js Serv­er SDK (raw audio access for record­ing and mod­er­a­tion bots; option­al end-to-end encryp­tion). https://docs.4players.io

[11] 4Players on GitHub: odin-unreal-wwise-adapter‑1.x. https://github.com/4Players/odin-unreal-wwise-adapter‑1.x

[12] 4Players: ODIN Cor­tex, still in ear­ly access as of 1 Sept 2026 ("apply to test Cor­tex on your project ahead of gen­er­al avail­abil­i­ty"). https://odin.4players.io/cortex

[13] Reg­u­la­tion (EU) ²⁰²²⁄₂₀₆₅ (Dig­i­tal Ser­vices Act), Art. 16(2). https://eur-lex.europa.eu/eli/reg/2022/2065/oj

[14] Sec­tion 201 of the Ger­man Crim­i­nal Code (StGB). https://www.gesetze-im-internet.de/stgb/__201.html

[15] Neg­a­tive find­ing on case law and legal lit­er­a­ture: own research and two LLM-assist­ed ver­i­fi­ca­tion search­es in freely acces­si­ble data­bas­es (open­Jur, dejure, rewis, law-firm pub­li­ca­tions) in August 2026; beck-online and juris not searched. Addi­tion­al­ly, Research Ser­vices of the Ger­man Bun­destag, WD 7 - 3000 - ⁰⁸¹⁄₂₁ of 30 July 2021, p. 5, fn. 12: "No crim­i­nal court deci­sions on this ques­tion are appar­ent" (author's trans­la­tion); the sen­tence there refers to self-record­ed voice mes­sages in mes­sen­gers, not to voice chat.

[16] Ger­man Fed­er­al Court of Jus­tice (BGH), judg­ment of 13 Oct 1987, VI ZR ⁸³⁄₈₇ (NJW 1988, 1016): a gen­er­al pri­vate inter­est in obtain­ing evi­dence about the con­tent of a con­ver­sa­tion does not jus­ti­fy the intru­sion into per­son­al­i­ty rights; spe­cial cir­cum­stances are required. Civ­il deci­sion on the admis­si­bil­i­ty of covert record­ings, not on Sec­tion 201 StGB. https://dejure.org/dienste/vernetzung/rechtsprechung?Gericht=BGH&Datum=13.10.1987&Aktenzeichen=VI+ZR+83/87. Applied to a covert record­ing of a work­place con­ver­sa­tion: Region­al Labour Court Rhineland-Palati­nate, 19 Nov 2021, 2 Sa ⁴⁰⁄₂₁, https://dejure.org/dienste/vernetzung/rechtsprechung?Gericht=LAG+Rheinland-Pfalz&Datum=19.11.2021&Aktenzeichen=2+Sa+40/21; there also Fed­er­al Labour Court (BAG), 19 July 2012, 2 AZR ⁹⁸⁹⁄₁₁: the non-pub­lic spo­ken word of anoth­er may not be covert­ly record­ed, even at work.

[17] Embark Stu­dios, Help Cen­ter: In-game com­mu­ni­ca­tion (Raider Voice as opt-in, selec­table for par­ty and/or prox­im­i­ty). https://id.embark.games/arc-raiders/support/faq/159-in-game-communication. Ten pre­sets, five male and five female voic­es: Embark gives no offi­cial fig­ure; the split is described in Del­ti­a's Gam­ing, Arc Raiders: How To Use Voice Chang­er, 4 Nov 2025, https://deltiasgaming.com/arc-raiders-how-to-use-voice-changer/.

[18] Pock­et Tac­tics: inter­view with design direc­tor Vir­gil Watkins, Novem­ber 2025. https://www.pockettactics.com/arc-raiders/interview-proximity-chat

[19] Epic Games, Unre­al Engine devel­op­er inter­view: Embark Stu­dios build the award-win­ning ARC Raiders with Unre­al Engine, March 2026 (Epic Online Ser­vices Voice; cus­tom voice-chang­er tech­nol­o­gy; "safe space" in the arti­cle text). https://www.unrealengine.com/developer-interviews/embark-studios-build-the-award-winning-arc-raiders-with-unreal-engine

[20] Hil­len­brand, J. M. / Clark, M. J. (2009): The role of f0 and for­mant fre­quen­cies in dis­tin­guish­ing the voic­es of men and women. Atten­tion, Per­cep­tion & Psy­chophysics 71(5), 1150-1166. https://doi.org/10.3758/APP.71.5.1150. Also Skuk, V. G. / Schwein­berg­er, S. R. (2014): Jour­nal of Speech, Lan­guage, and Hear­ing Research 57(1), 285-296 (fun­da­men­tal fre­quen­cy as the strongest sin­gle cue; joint mor­ph­ing of fun­da­men­tal fre­quen­cy and for­mants pro­duces changes in per­ceived gen­der cat­e­go­ry equiv­a­lent to mor­ph­ing all para­me­ters). https://doi.org/10.1044/1092-4388(2013/12-0314)

[21] Cote, A. C. (2017): "I Can Defend Myself": Wom­en's Strate­gies for Cop­ing With Harass­ment While Gam­ing Online. Games and Cul­ture 12(2). https://doi.org/10.1177/1555412015587603. Fox, J. / Tang, W. Y. (2017): Wom­en's expe­ri­ences with gen­er­al and sex­u­al harass­ment in online video games. New Media & Soci­ety 19(8) ("gen­der bend­ing or gen­der neu­tral­iza­tion through screen name or avatar choice, avoid­ing com­mu­ni­ca­tion with oth­er play­ers"). https://doi.org/10.1177/1461444816635778. Nei­ther abstract names the voice explic­it­ly. The voice chang­er as a strat­e­gy is named in Deng, Z. (2024/2026), New Media & Soci­ety 28(1), 270-288, https://doi.org/10.1177/14614448241287831, and Crothers, H. / Scott-Brown, K. C. / Cun­ning­ham, S. J. (2024/2026), Games and Cul­ture 21(4), 746-764, https://doi.org/10.1177/15554120241273358. The VR find­ing: Povinel­li, K. C. / Zhao, Y. (2024), Spring­board, Road­block or "Crutch"?, IEEE VR 2024, 354-364, https://doi.org/10.1109/VR58804.2024.00057 (inter­views with 13 trans and gen­der-non­con­form­ing users, qual­i­ta­tive, not an effi­ca­cy test). Kao, D. et al. (2024): Explor­ing how gen­der-anony­mous voice avatars influ­ence wom­en's per­for­mance in online com­put­ing group work. Inter­na­tion­al Jour­nal of Human-Com­put­er Stud­ies 181, 103146. https://doi.org/10.1016/j.ijhcs.2023.103146

[22] Björn Burghard: A Foy­er and a Tram in Brno, and the Gaps In Between. Way­cleft, June 2026. https://waycleft.com/gameaccess-brno-the-gaps-in-between/

[23] Games­Beat (Dean Taka­hashi): Inside ARC Raiders. A can­did inter­view with game lead­ers Patrick Söder­lund and Owen Mahoney while play­ing live, 30 Dec 2025. The descrip­tion of the archi­tec­ture ("archi­tect­ed to cre­ate ten­sion") is the arti­cle's own text. https://gamesbeat.com/inside-arc-raiders-a-candid-interview-with-game-leaders-patrick-soderlund-and-owen-mahoney-while-playing-live/

[24] MADFINGER Games: Devlog 5 on the Night Ops update, 27 Nov 2024 ("We're final­ly intro­duc­ing an in-game report­ing sys­tem"), https://www.madfingergames.com/news/devlog5. Bug or Play­er Report­ing, pinned stu­dio post in the Steam forums for Gray Zone War­fare, 10 Dec 2024 (reports of cheat­ing and harass­ment by tick­et via support.madfingergames.com; "To help us take action, we kind­ly request that you pro­vide proof"), https://steamcommunity.com/app/2479810/discussions/0/592882992704238535/. End User License Agree­ment, https://www.madfingergames.com/eula, code-of-con­duct sec­tion: hate speech, dis­crim­i­na­to­ry lan­guage, harass­ment and threats lead to tem­po­rary or per­ma­nent bans; play­ers are encour­aged to report; false report­ing can be pun­ished. The doc­u­ment applies to all MADFINGER games and con­tains pas­sages from the mobile era; it says noth­ing about who han­dles reports or what is pre­served from voice inci­dents.

[25] Epic Online Ser­vices, tech­ni­cal blog: More details on how we built voice report­ing (Fort­nite, Rock­et League, Fall Guys; buffer of the last five min­utes on devices, dig­i­tal­ly signed, upload only on an active report, no con­tin­u­ous serv­er-side record­ing; serv­er-side stor­age of the sub­mit­ted clip after a report; buffer­ing by the EOS SDK). No com­pa­ra­ble voice-report­ing fea­ture for exter­nal licensees is doc­u­ment­ed in EOS release notes or the pub­lic API ref­er­ence of the past twelve months; a gener­ic audio-buffer inter­face (EOS_RTCAudio_AudioBuffer) is avail­able. https://onlineservices.epicgames.com/en-US/news/technical-blog-more-details-on-how-we-built-voice-reporting. Announce­ment 16 Nov 2023: https://www.epicgames.com/site/en-US/news/introducing-voice-reporting-in-fortnite

[26] Facepunch Stu­dios, Sup­port: Mut­ing and Cen­sor­ship (as of 23 Aug 2024; mut­ing cov­ers text and voice at once; the word fil­ters men­tioned there are Steam's glob­al chat fil­ter­ing, the cen­sor­ship options con­cern visu­al con­tent), https://support.facepunchstudios.com/hc/en-us/articles/10641641796253-Muting-and-Censorship; Report­ing in Rust (F7, death screen, con­tact card, chat; note on com­mu­ni­ty servers' own report­ing chan­nels), https://support.facepunchstudios.com/hc/en-us/articles/10640775389853-Reporting-in-Rust; Banned for Racism/Homophobia (as of 11 June 2026: hate and slurs on Facepunch servers are con­tin­u­ous­ly mon­i­tored and auto­mat­i­cal­ly flagged to staff, then man­u­al­ly reviewed; bans per­ma­nent and not appeal­able; con­tin­ued play on com­mu­ni­ty servers pos­si­ble; the page does not dis­tin­guish text from voice), https://support.facepunchstudios.com/hc/en-us/articles/360007485418-Banned-for-Racism-Homophobia. Addi­tion­al­ly: Facepunch, Com­mu­ni­ty Serv­er and Host­ing Guide­lines, as of 15 July 2025, which assign respon­si­bil­i­ty for servers and for han­dling rule vio­la­tions to their oper­a­tors; vio­la­tions of the guide­lines can lead to removal from the serv­er list or fur­ther sanc­tions; no spe­cif­ic state­ment on record­ing or auto­mat­ed eval­u­a­tion of voice chat. https://facepunch.com/legal/servers

[27] Riot Games Pri­va­cy Notice (as of 28 April 2026), sec­tion I.A (legal basis "Legit­i­mate Inter­est" for "Play in a safe and respect­ful envi­ron­ment" and "Chat with oth­er play­ers"; con­sent only for mar­ket­ing), sec­tion I.D ("where legal­ly per­mis­si­ble"), sec­tion III (reten­tion), sec­tion VI.A (right to object). https://www.riotgames.com/en/privacy-notice. Statu­to­ry basis: Art. 6(1)(f), Art. 21 GDPR.

[28] Uni­ty Tech­nolo­gies / The Har­ris Poll: Tox­i­c­i­ty in Mul­ti­play­er Games Report 2023. Sur­veyed from 21 to 31 July 2023: 2,522 play­ers (US 1,046, UK 633, South Korea 843); from 21 July to 9 Aug 2023: 407 devel­op­ers (US 202, UK 101, South Korea 104); pre­ci­sion ±2.7 per­cent­age points for play­ers, ±5.4 for devel­op­ers. Val­ues used: 96 per­cent of devel­op­ers con­sid­er effec­tive tools avail­able, 52 per­cent name invest­ment in them a top pri­or­i­ty for their stu­dio (Find­ing 4); 56 per­cent admit that pro­tec­tion from tox­ic behav­iour is not a top pri­or­i­ty, 89 per­cent con­sid­er more solu­tions nec­es­sary (p. 14); report­ing func­tions imple­ment­ed at 54 per­cent and bans at 47 per­cent (p. 17); 68 per­cent of play­ers are will­ing to be record­ed for mod­er­a­tion pur­pos­es; the report also describes gen­der dif­fer­ences in will­ing­ness to be record­ed (p. 11). Avail­able via https://create.unity.com/toxicity-report (reg­is­tra­tion required).

[29] IGN (Michael Cripe): The Big Arc Raiders Inter­view: Embark Boss Patrick Söder­lund Talks Live-Ser­vice Plans, Ban­ning Cheaters, and the Secret Sauce of PvE, 5 Jan 2026 (Söder­lund on the unpre­dictabil­i­ty of fel­low play­ers: "It is just the nature of a mul­ti­play­er game"). https://www.ign.com/articles/the-big-arc-raiders-interview-embark-boss-patrick-soderlund-talks-live-service-plans-banning-cheaters-and-the-secret-sauce-of-pve

[30] Kociel­nik, R. / Li, Z. / Line­gar, M. / Sam­bra­no, D. / Soltani, F. / Kim, M. / Naqvie, N. / Cahill, G. / Anand­ku­mar, A. / Alvarez, R. M. (2025): Online Mod­er­a­tion in Com­pet­i­tive Action Games: How Inter­ven­tion Affects Play­er Behav­iors. Pro­ceed­ings of the ACM on Human-Com­put­er Inter­ac­tion 9(6), 96-130. https://doi.org/10.1145/3748599. Qua­si-exper­i­men­tal design on Call of Duty: Mod­ern War­fare II, includ­ing voice vio­la­tions; preprint since 1 Nov 2024. Peer-reviewed, but not an inde­pen­dent exter­nal eval­u­a­tion: data and sev­er­al authors come from the com­pa­ny's envi­ron­ment.

[31] Games­Radar+: Arc Raiders devs "still learn­ing" what to do with PvP, but Embark design direc­tor says PvE fights are doing their job, 5 May 2026. Design direc­tor Vir­gil Watkins names unpre­dictabil­i­ty as a val­ued qual­i­ty: "You're not just design­ing around AI encoun­ters or play­er-ver­sus-play­er com­pe­ti­tion in iso­la­tion but rather cre­at­ing sit­u­a­tions where those sys­tems inter­act." https://www.gamesradar.com/games/third-person-shooter/arc-raiders-devs-still-learning-what-to-do-with-pvp-but-embark-design-director-says-pve-fights-are-doing-their-job-they-make-the-player-interactions-even-more-impactful/

[32] Vera-Gray, F. / Kel­ly, L. (2020): Con­test­ed gen­dered space. Pub­lic sex­u­al harass­ment and wom­en's safe­ty work. Inter­na­tion­al Jour­nal of Com­par­a­tive and Applied Crim­i­nal Jus­tice 44(4), https://doi.org/10.1080/01924036.2020.1732435; iden­ti­cal as a book chap­ter, https://doi.org/10.4324/9780429352775-15, full text at https://www.sakraplatser.abe.kth.se/wp-content/uploads/sites/60/2020/07/10.4324_9780429352775-15.pdf. The term goes back to Kel­ly (2012). Find­ings used: safe­ty work is pre-emp­tive, invis­i­ble labour per­formed "just in case", before any­thing hap­pens; the mes­sage that accu­mu­lates from it over a life­time the authors describe as a demand to be less, express­ly includ­ing "less vocal, less vis­i­ble, less free"; the work turns from an option into a require­ment, some­thing a woman is rather than does, and who­ev­er fails to per­form it is seen as being wrong rather than hav­ing done some­thing wrong; the suc­cess of the work is an absence and there­fore can­not be shown, only the fail­ures can be count­ed, which the authors describe as the catch-22 of the "right amount of pan­ic"; as one form of the work they name strate­gic alien­ation, the erect­ing of bar­ri­ers such as head­phones or sun­glass­es between one­self and the sur­round­ings. The sub­ject of the study is phys­i­cal pub­lic space; the trans­fer to voice chan­nels in games is the author's, not theirs.

[33] Björn Burghard: Two in the Morn­ing, No Call. Way­cleft, 7 June 2026. https://waycleft.com/two-in-the-morning-no-call/ (Ger­man ver­sion: https://waycleft.com/de/nachts-um-zwei-ohne-anruf/). Safe­ty work is intro­duced there, with the same sources and with a sur­vey on avoid­ing dark places.

[34] Mod­u­late: Tox­Mod, prod­uct page (live mon­i­tor­ing, detec­tion while it hap­pens, warn­ings and auto­mat­ed actions in real time, eval­u­a­tion of tone, inten­si­ty and esca­la­tion pat­terns), https://www.modulate.ai/products/toxmod. Ven­dor's own claims. Three-stage esca­la­tion with shad­ow mute at stage two: Modulate/Activision case study [7].

[35] On com­bin­ing voice, text and game­play data: Getgud.io adver­tis­es a serv­er-side plat­form that eval­u­ates text chat, voice via tran­scripts and game­play actions (posi­tions, kills, team-kills, spawn camp­ing) togeth­er, https://www.getgud.io/blog/top-3-best-ai-chat-moderation-tools-for-games/; ven­dor's self-descrip­tion, the doc­u­ment­ed use cas­es con­cern cheat­ing and grief­ing. Mod­u­late's "risk cat­e­gories" eval­u­ate behav­iour trends over longer peri­ods, but only with­in speech (Sil­i­co­nAN­GLE, 31 Aug 2023, https://siliconangle.com/2023/08/31/activision-partners-modulate-combat-game-toxicity-ai-powered-voice-chat-moderation/). Sony patents describe mod­er­a­tion sys­tems that com­bine com­mu­ni­ca­tion data with con­text from game­play and fur­ther user char­ac­ter­is­tics: US 11,266,912 B2 (Sony Inter­ac­tive Enter­tain­ment LLC, grant­ed 8 March 2022) on detect­ing and pro­cess­ing dis­rup­tive play­er behav­iour, with game teleme­try, chat and video stream as review mate­r­i­al; US 12,148,445 B2 (Sony Inter­ac­tive Enter­tain­ment Inc., grant­ed 19 Nov 2024) on emo­tion-based mod­er­a­tion from com­mu­ni­ca­tion, game­play and phys­i­o­log­i­cal sig­nals. A patent is not evi­dence of prod­uct intent. Research back­ground: Yehor Tereshchenko and Mika Hämäläi­nen (2025): "Effi­cient Tox­i­c­i­ty Detec­tion in Gam­ing Chats: A Com­par­a­tive Study of Embed­dings, Fine-Tuned Trans­form­ers and LLMs", arX­iv preprint; sec­tion 7.2.2 sketch­es research direc­tions for mod­er­at­ing text, audio and image/video con­tent. The paper does not exam­ine the joint eval­u­a­tion of speech con­tent and pur­suit move­ments in games and does not doc­u­ment a spe­cif­ic research gap for that. https://arxiv.org/html/2510.17924v1