PlayStation Patented AI Decoy Accounts for PSN Safety

Sony Interactive Entertainment was granted a patent this month for a "Large Language Model Powered Social Integrity System." The idea is simple and a little unsettling: PlayStation Network would run fake accounts, controlled by an LLM, that look like real players. Suspicious users who approach those accounts get flagged before they reach an actual person.

The filing (application 18/314775, issued as US 12,706,952 on August 11, 2026) was published in coverage on August 13. Sony has not announced this as a live PSN feature. Patents describe what a company wants the option to build, not what ships next firmware update. It still tells you where platform moderation is headed.

How the decoy is supposed to work

The patent calls the bot a "pseudo account." It is set up to mimic a target demographic's messaging habits and to appear legitimate to someone looking for victims. The filing is explicit that this can include mimicking the messaging behavior of minors and placing that behavior where "anti-social" users hunt for it.

When a suspicious account contacts the decoy, the system treats that as a signal, not an automatic ban. The flow described in the abstract:

The pseudo account is placed where risky users look.
That user starts a conversation with the AI-controlled profile.
The account is flagged as suspicious.
The system can wait until the same user approaches a real player, then warn or intercept before the harmful message lands.

Behaviors the system is designed to watch for include moving a chat off-platform, asking for personal information, contacting accounts that appear to belong to minors, scam attempts, spam, and cheating-related activity. The LLM in the filing is described in GPT-style terms — a generative model running the decoy's side of the conversation.

Why this is not an instant ban hammer

Sony's design, on paper, is conservative about punishment. Talking to the decoy is evidence, not a conviction. The goal is to mark the account so that the next time it reaches a real kid, a real player, or a potential scam target, PSN can warn or block first.

That is more proactive than today's report-and-wait model, where the harm often happens before a moderator sees the ticket. It also raises the obvious accuracy problem. A system that warns about genuine predators makes PSN safer. A system that flags awkward social players, joke accounts, or anyone who messaged the wrong profile is a new kind of false positive — one the reported-against person may never see.

Takeaway: This is a patent, not a product announcement. The intent is to catch scammers and predators before they reach real players. The open questions are accuracy, transparency, and whether anyone who gets flagged will ever know they were talking to a bot.