Activision has been quietly publishing some of the most detailed community safety data in all of gaming since August 2023.
Detailed anti-toxicity progress reports. Real numbers. Before-and-after comparisons. Methodology notes. Published openly on the Call of Duty blog for anyone to read.
Almost no one has.
I’ve spent time going through every report they’ve published, cross-referencing the numbers, and trying to figure out what they actually mean for the average person playing Call of Duty on a Tuesday night. What I found is more interesting than I expected. The data shows real, measurable progress, but it also raises questions that the reports don’t answer. And the fact that Call of Duty is the only major franchise publishing this level of detail says something uncomfortable about the rest of the industry.
Let me walk you through it.
A Little Background: Why CoD Is the Case Study
Call of Duty partnered with a company called Modulate to deploy their AI voice moderation tool, ToxMod, starting in August 2023. First as a beta in Modern Warfare II, then going fully global (excluding Asia) with Modern Warfare III in November 2023, and continuing through Black Ops 6.
ToxMod listens to voice chat in real time, analyzes tone and content using a multi-layered AI model, and flags potential violations for human moderator review. No automated bans. A person makes every enforcement decision. If you want a deeper explanation of how the technology works, we broke that down in a separate piece.
The important thing for this article is that Activision committed to transparency about the results. They said they’d publish data on how it was performing, and they’ve actually done it. Three major reports over roughly two years, with updated numbers each time.
For a gaming industry that usually treats moderation data like a state secret, that’s genuinely notable.
The Numbers, Straight
Let me give you the raw data first before I tell you what to make of it.
From the first report (August 2023, covering the MW2 beta period):
- 50% reduction in toxicity exposure in voice chat after ToxMod deployment
From the October 2024 report (covering the period leading into Black Ops 6):
- 43% drop in exposure to toxic voice chat compared to the pre-ToxMod baseline
- 67% reduction in repeat offenders since June 2024
- 80% of players who received a voice chat enforcement did not reoffend
- 45 million text messages blocked across 20 languages
- Only 23% of player-generated reports contained usable evidence
That last number deserves a moment. Out of every 100 reports a player files, 77 of them don’t give a human moderator anything to work with. That’s not a knock on players filing reports, it reflects how hard it is to capture clear evidence of voice chat harassment without a system that’s already recording and flagging it. The AI moderation is running parallel to the report system, not dependent on it. But it tells you how limited the old “just press report” model always was.
What These Numbers Actually Mean
Let me translate each of these into plain language, because percentages without context can mean a lot of things.
The 50% reduction in toxicity exposure. This was the beta result, and it’s the flashiest number. What it means in practice: a player who would have had, say, 10 sessions per week with a toxic voice chat incident could now expect about 5. That’s not nothing. That’s the difference between something that ruins your evening once a week versus twice a week. Over time, that compounds.
The caveat worth noting: this was a beta period, which means a smaller, more controlled rollout. Beta numbers often look better than full-launch numbers because you’re starting with a more engaged, less anonymous player base. The fact that the sustained number is 43% rather than 50% suggests some of that is true, but 43% is still a real and significant reduction.
The 43% sustained drop. This one matters more than the beta number because it’s what held over two full game releases. Moderation systems often see initial improvements that fade as bad actors adapt. A 43% sustained reduction across MW2 and into Black Ops 6 suggests the system isn’t being gamed as easily as keyword filters used to be. The AI is doing something harder to circumvent than a word list.
The 67% drop in repeat offenders. This is the number I find most interesting. The goal of any enforcement system isn’t just to remove bad behavior in the moment, it’s to change behavior over time. A 67% reduction in repeat offenders since June 2024 means that the majority of players who got flagged and received enforcement didn’t come back and do it again.
There are two ways to read that. The optimistic read: enforcement worked, people learned, behavior changed. The skeptical read: some repeat offenders just made new accounts. Activision doesn’t break down how many of the non-repeats are reformed players versus banned accounts that simply moved on. Both are probably true to some extent.
The 80% non-reoffend rate. Related to the above, but this is the per-player version. Eight out of ten players who received enforcement didn’t get flagged again. That’s a better outcome than most punishment-based systems achieve, in gaming or anywhere else.
45 million text messages blocked. This is the text moderation number, powered by Community Sift (the Microsoft-owned platform Two Hat built). It’s a big number, but it’s also hard to contextualize without knowing what percentage of total messages it represents. Call of Duty has hundreds of millions of active players globally. 45 million blocked messages could be a lot or a little, depending on the denominator, which Activision doesn’t publish.
The Things the Reports Don’t Tell You
Good data analysis means asking what’s missing, not just what’s there. The CoD reports are genuinely good by industry standards, but there are gaps worth naming.
No absolute baseline numbers. Every metric is a percentage change. We know toxicity exposure dropped 43%, but we don’t know what the starting point was. Was it 10 incidents per 100 sessions? 30? 50? Without the baseline, it’s hard to know whether a 43% reduction means you now have a pleasant experience or a significantly-improved-but-still-bad experience.
No breakdown by game mode. The numbers are aggregated across all of Call of Duty. Competitive ranked modes almost certainly have very different toxicity profiles than casual playlists or Warzone. Knowing whether the AI is more effective in one context versus another would be genuinely useful for players deciding where to spend their time.
No regional data. ToxMod doesn’t cover Asia. The reports don’t say much about how results vary across North America, Europe, and other covered regions. Given that cultural context affects how tone and language are interpreted by AI systems, this would be valuable to understand.
No false positive rate. How often does the system flag something that turns out not to be a violation? Ubisoft publishes this number for Rainbow Six Siege (0.1%). Activision doesn’t. Given that the entire enforcement chain includes human review specifically to catch false positives, this should be knowable and publishable.
The appeals data is thin. Activision’s support site offers ban appeals, but the reports don’t include data on how many appeals are filed, how many succeed, or how long the process takes. For a system where a human makes every enforcement decision, some transparency on error rates and corrections would strengthen the credibility of the whole operation.
None of this invalidates what’s there. It just means the picture is incomplete in ways that matter.
Grading Activision’s Transparency
I want to be fair here. The fact that these reports exist at all puts Activision ahead of most of the industry. Let me actually compare.
Xbox publishes semi-annual transparency reports covering the entire platform, enforcement volumes, proactive-vs-reactive breakdown, tool innovation details, child safety referrals. Their 2026 report covering 2025 included data on 14.8 billion pieces of content moderated. It’s the gold standard in platform-level gaming transparency.
EA published its first transparency report in 2024, covering FIFA, Madden, and other EA titles. It includes data on texts scanned and penalties issued. Less detailed than Xbox, but a real start.
Riot Games publishes behavioral system updates periodically on their dev blogs, with data on how many penalties were issued and recidivism rates. Not a formal annual report, but real numbers appear.
Ubisoft publishes its Player Safety Transparency Report, which includes false positive rates. Rare that a studio publishes that.
PlayStation published its first Online Safety Report in February 2025, nearly a decade after Xbox started the practice. The data is much thinner than Xbox’s.
Nintendo publishes nothing. No moderation data, no transparency reports, no enforcement statistics. The most opaque major gaming platform in the industry.
Call of Duty‘s reports sit in the upper tier of this landscape, better than most game-specific publishers, less comprehensive than Xbox’s platform-level reports. The game-specific data is more useful for players of that specific franchise than a platform-wide report, because you know exactly what you’re getting when you load up Warzone.
My grade: B+. Meaningful data, real commitment to publishing it, docked for missing baseline numbers and false positive rates.
The Bigger Takeaway
Here’s the thing that stuck with me after going through all of this.
Call of Duty has a genuinely rough reputation for community toxicity. It’s the game you think of when someone says “terrible voice chat.” The lobby screamer, the post-game harassment, the endless slurs from anonymous accounts.
And Activision, by publishing this data, is effectively saying: we know the reputation, we’re trying to change it, and here’s evidence that we’re making progress.
The 43% reduction in toxic voice chat exposure isn’t a fixed game. It’s a game that was broken and is getting less broken. That’s different from a game that was always fine. It’s actually harder to do.
What would make this story complete is if other studios followed. If Riot published voice moderation data the way they publish behavioral system updates. If Epic published Fortnite toxicity metrics. If Blizzard published Overwatch 2 numbers. If every major multiplayer game had a document you could point to and say, “here’s what they’re doing and whether it’s working.”
Right now, one franchise publishes this, and the rest don’t. That gap is worth paying attention to. Because the studios not publishing data aren’t necessarily doing less, they might just be less transparent about what they’re doing. But you can only hold companies accountable for what they’re willing to show you.
Activision is showing you something. That’s worth acknowledging, even if you also want to push for more.
What This Means If You Play Call of Duty
Practically speaking, here’s what the data tells you about your actual experience:
Voice chat has gotten meaningfully safer over the past two years. Not perfect. But a 43% reduction in exposure to toxic voice chat is real, and the sustained repeat-offender reduction suggests it’s not temporary.
Reporting still matters, but it works differently than most players assume. When you report someone in CoD, you’re not directly causing their enforcement. You’re adding a data point to their behavioral record that the AI moderation system uses alongside its own detections. Think of it less like calling the police and more like adding to a pattern file.
Text chat is being moderated at a massive scale. 45 million blocked messages means the text chat filter is running aggressively. If you’ve noticed text toxicity feeling lower in recent CoD titles than older ones, that’s likely a real difference, not a perception shift.
The system isn’t perfect, and false positives happen. If you receive an enforcement action you believe is wrong, appeal it. The appeal process exists specifically for this. Human review means human error, and the process is designed to catch that.
And if you’re on the fence about whether any of this matters, whether publishing data and deploying AI moderation is PR or something real, the 80% non-reoffend rate is the most honest answer I can give you. Four out of five people who got caught, didn’t do it again. That’s the system working.
Read something in those reports that I missed? Disagree with my take on the data? I want to hear it. Drop your thoughts in the comments or find me in the TAG Discord. This is an ongoing series and player perspectives on what the numbers mean in practice are how it gets better.
