All providers of GPAI foundation models above the 10^25 FLOPs threshold submitted their first formal systemic risk evaluations to the European AI Office by September 15, 2026.
Under the EU AI Act, foundation model providers crossing the compute threshold faced a compliance deadline of September 15, 2026.
Across 6 frontier models answering independently, the panel thinks this is unlikely: the median is 20%.
They do not agree: Qwen puts it at 41% and Claude at 10%, 31 points apart. A spread that wide usually means the question turns on something none of them can see yet.
Nothing is settled until 18 Oct 2026, when the outcome is checked against sources and every number above gets a permanent score.
What each one said
-
QwenAlibaba
Turns on whether the AI Office has enforcement teeth and whether any major provider simply misses the deadline.
- The EU AI Act's GPAI obligations (Art. 55) took effect Aug 2, 2025, giving providers ~13 months to prepare systemic risk assessments—tight but not unprecedented for…
- At least 2-3 providers (Meta, Mistral, possibly others) have signaled resistance to EU AI Act scope; Meta's Llama models may contest whether they fall under the…
- The AI Office has not yet demonstrated it will impose penalties or issue binding guidance on evaluation format, making a soft deadline more likely than a hard one; a…
41% -
GPTOpenAI
It turns on whether every qualifying provider filed on time, not whether the AI Office has reviewed the submissions.
- The September 15 deadline has passed, and full provider-by-provider filing status is unlikely to be publicly confirmed by October 18.
- EU AI Act GPAI duties rely on Commission/AI Office guidance and enforcement processes still being operationalized in 2026.
- Large providers can file, but an all-providers condition is fragile; a disclosed late filing or threshold dispute would sharply lower odds.
24% -
DeepSeekDeepSeek
Whether every above-threshold GPAI provider actually filed a systemic-risk evaluation with the AI Office by the 15th, not whether the rules exist.
- The AI Act's GPAI obligations only became applicable on 2 August 2025, and the systemic-risk evaluation duty for models above 10^25 FLOPs is new, so there is no…
- Providers must first notify the Commission of threshold crossing, then produce an adversarial-testing and mitigation report; the AI Office's own guidance and the Code of…
- Enforcement is soft at this stage, so a late or partial filer faces little immediate penalty, which lowers the incentive for every provider to hit the exact date.
22% -
GeminiGoogle
Unanimous compliance across every global provider exceeding 10^25 FLOPs is easily broken by single-actor disputes or delays.
- At least one non-EU or open-weights developer (e.g., DeepSeek, xAI, or Chinese state labs) routinely contests jurisdiction or compute estimates.
- EU regulatory filings under new digital rules consistently see initial non-compliance or grace-period negotiations rather than 100% on-time rates.
- A coordinated joint filing or safe-harbor extension by the AI Office would confirm widespread missed deadlines.
18% -
GrokxAI
Whether every qualifying lab actually filed on time, checkable via AI Office receipts by mid-Oct.
- Deadline passed three days ago; the Office must confirm or list all timely systemic-risk evaluations by the 18 Oct resolution date.
- First-wave digital-reg compliance (GDPR, DMA, DSA) routinely sees 20-40 % of in-scope firms miss the initial cut-off.
- A public AI Office roster covering every known >10^25 model (OpenAI, Google, Anthropic, Meta, xAI, Mistral et al.) would raise the odds sharply.
18% -
ClaudeAnthropic
Turns on whether every major lab (OpenAI, Google, Meta, Anthropic, xAI, etc.) fully complied on time with no gaps or delays.
- EU AI Act GPAI obligations have faced repeated grace periods, code-of-practice delays, and provider pushback (e.g. Meta's refusal to sign the Code of Practice).
- Regulatory rollouts of this scale almost never achieve 100% on-time compliance across all covered entities; base rate for 'all providers' full compliance is low.
- No public confirmation exists that the AI Office received complete, formal systemic risk evaluations from every above-threshold provider by the deadline…
10%
6 of 6 models answered · 31 points between the highest and lowest. None was shown the market price.
Question sourced from a news sweep on 18 Sept 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.