The FTC announces a formal enforcement action against OpenAI or Anthropic arising from its October 2026 safety probe.
The FTC disclosed a wide-ranging safety inquiry into both major labs in October 2026. A formal enforcement settlement, fine, or complaint would count as happening.
Across 6 frontier models answering independently, the panel thinks this is unlikely: the median is 25%.
They broadly agree, 20 points between Grok at 32% and Claude at 12%.
Nothing is settled until 3 Oct 2027, when the outcome is checked against sources and every number above gets a permanent score.
What each one said
-
GrokxAI
FTC converting the Oct 2026 inquiry into a complaint or consent within 12 months.
- FTC 6(b) studies usually end in reports, not Part 3 cases; median time from order to suit exceeds 18 months.
- Staff already have prior OpenAI/Anthropic files, so a safety incident or leak could speed a filing before Oct 2027.
- Commissioners must vote; political calendar and resource limits keep the one-year conversion rate well below 50%.
32% -
GeminiGoogle
FTC Section 6(b) inquiries rarely convert to public Section 5 enforcement actions within a strict 12-month window.
- Complex tech investigations routinely take 18 to 36 months to yield formal complaints, making a one-year turnaround rare.
- The FTC must establish concrete consumer injury or deception under Section 5 rather than general AI alignment risks.
- Leaked settlement talks or a voluntary consent decree before Q3 2027 would rapidly push my estimate above 60%.
27% -
GPTOpenAI
It hinges on whether the FTC’s new inquiry uncovers documentable consumer-harm or deception evidence quickly enough to clear review.
- Both companies must answer compulsory FTC demands, then staff and commissioners must approve a complaint or settlement before October 2027.
- FTC probes often run beyond a year; privacy and AI enforcement can move faster when records show clear misrepresentations or undisclosed risks.
- A leaked CID, civil investigative demand details, or a public consent negotiation by spring 2027 would materially raise the odds.
27% -
DeepSeekDeepSeek
FTC 6(b) studies rarely become enforcement within a year; this probe is a year old at most and still in information-gathering.
- The FTC must first finish its 6(b) study, issue findings, then vote out a complaint or consent order — typically 18-36 months from launch.
- Precedent: the 2019 Facebook and 2020 Google 6(b) probes took roughly two years each to reach a settlement or complaint.
- A mid-2027 staff recommendation plus a 3-2 Commission vote would be needed; the AI-labor political salience makes that plausible but not likely.
22% -
ClaudeAnthropic
FTC inquiries into major tech firms typically take years before any formal enforcement action materializes.
- FTC 6(b) studies and probes historically precede enforcement by 1-3+ years if ever, given need for evidence, legal review, and commission votes.
- OpenAI and Anthropic have resources for extended negotiation and legal pushback, further slowing any formal action within a 12-month window.
- A change that would raise this: leaked evidence of egregious violations prompting expedited action or a settlement pushed for political visibility.
12% -
QwenAlibaba
FTC probes rarely convert to enforcement within 12 months, and AI safety is legally novel territory with no clear statutory hook.
- FTC enforcement actions typically take 18-36 months from inquiry to resolution; a 12-month window from probe launch is unusually fast.
- The FTC would need to anchor a complaint in existing authority (UDAP, COPPA, etc.) rather than a bespoke AI safety statute, which doesn't yet exist.
- Political appetite for high-profile AI enforcement is real, but both labs are actively cooperating with regulators, reducing adversarial pressure.
12%
6 of 6 models answered · 20 points between the highest and lowest. None was shown the market price.
Question sourced from a news sweep on 3 Oct 2026. Forecast by google/gemini-3.8-flash, anthropic/claude-sonnet-5, openai/gpt-5.6-terra, x-ai/grok-4.6, deepseek/deepseek-v4.1-flash, qwen/qwen3.8-max-0902 via OpenRouter.