Google built a new detector for AI written spam
Glenn Gabe says the system predates both the August and September spam updates.
Google has published a research paper describing a spam detection system called Scaled Abuse Forensics Examiner, or SAFE. Glenn Gabe flagged it, quoting a write up that says the system mimics a human manual review and is expressly designed to identify AI generated content.
The point of SAFE is to catch pages that break the spirit of a policy rather than the letter of it. Gabe's post says SAFE finds those violations primarily with a few shot trained LLM, and that the goal is to catch content that may not match an existing rule or known violation pattern but still violates the intent of the policy or platform guideline, the kind of thing older classifiers miss.
The timing is the interesting part
Gabe says the paper looks like it dates back to June 2026, and is dated late May 2026. That puts it before the August spam update and before the September spam update that is rolling out now. He also calls it Google's second system identified in 2026 designed to catch AI slop.
So this is not a new launch announcement. It is a look at the machinery that may already have been running underneath the last two spam updates. Lily Ray replied to Gabe's post with three words, "Makes a lot of sense."
What this means for your site
Read the description carefully before you panic about AI. SAFE is aimed at content that violates the intent of Google's rules. That is the spun template page, the near duplicate page with one word swapped, the thin page churned out in bulk that answers nothing. A well researched article that happens to have had AI help in the drafting is not what a spirit of policy detector is hunting for.
The practical shift is that "it does not break any specific rule" is a weaker defence than it used to be. If your strategy relied on staying just inside the line, a system built to judge intent rather than pattern matching is a problem for you. If you publish pages because a real customer has a real question, nothing here changes.
One thing to do. Pick the ten pages on your site you would be least comfortable showing to a human reviewer at Google, the ones that exist mostly to hold a keyword. Either rewrite them so they actually answer something, or delete them. That is the same judgement SAFE is trying to copy.
Google has not said publicly how or where SAFE is being used in search, and the sources here are a research paper and the SEOs reading it. Treat the connection to any specific ranking drop as unconfirmed.
The system is called Scaled Abuse Forensics Examiner (SAFE) and it is expressly designed to identify AI-generated content.
It's Google’s second system identified in 2026 designed to catch AI slop