Provenance record

OpenAI axes next model citing safety issues

Financial Times: Artificial Intelligence (tier 1, news) 2026-09-29T00:48:02.000Z Original ↗

Source note: Headlines and standfirsts only; full text is paywalled.

Discourse valence

-7

Mixed or uncertain

confidence 65% · 4 items · range -100 to +65

-100

-50

+50

+100

-7

Adverse readingFavourable reading

Consensus of 4 models from different labs. Spread 165 points, agreement low.

Existential and catastrophic risk Loss of control and alignment Concentration of power

Excerpt as ingested

AI lab will hold back GPT-6.1 Astra following mounting pressure over misbehaviour by its agents

Every model that read this

Model	Provider	Stage	Score	Conf.	Latency	Prompt	When
Llama 3.3 70B	Meta	analysis	0	70%	2980ms	v1.0.0 / m1.0.1	2026-09-29 03:17
Claude Sonnet 5	Anthropic	consensus	+8	30%	7978ms	v1.0.0 / m1.0.1	2026-09-29 03:18
Gemini 2.5 Flash	Google	consensus	-100	80%	2017ms	v1.0.0 / m1.0.1	2026-09-29 03:18
Mistral Small 3.1 24B	Mistral AI	consensus	+65	80%	6238ms	v1.0.0 / m1.0.1	2026-09-29 03:18

Llama 3.3 70B · reading

Model held back due to safety issues.

evidence: reported horizon: n/a

Claude Sonnet 5 · reading

Headline-only report of OpenAI withholding a model over agent misbehaviour. Could reflect responsible governance, but also signals real problems with agent reliability. Too little detail to judge severity or scope, so implication stays near neutral with low confidence.

evidence: speculative horizon: n/a

Gemini 2.5 Flash · reading

OpenAI's decision to withhold a new model due to 'misbehaviour by its agents' strongly implies that the AI exhibited concerning, potentially harmful, behavior. This indicates a significant safety risk within advanced AI development. While the action taken (holding back the model) is positive, the underlying reason is materially concerning for humanity.

evidence: reported horizon: n/a

Mistral Small 3.1 24B · reading

The decision to hold back an AI model due to safety concerns shows a commitment to responsible AI development. This implies a positive step towards mitigating risks associated with advanced AI.

evidence: speculative horizon: n/a

Evidence extracted

OpenAI axes next model
GPT-6.1 Astra held back

The chain

SOURCE     Financial Times: Artificial Intelligence (tier 1)
   ↓
DOCUMENT   ed342d00-cad1-4922-b32e-86e621578000
           https://ft.com/content/488cb467-3cb7-4d06-9a5f-c0749c729d91
   ↓
EVIDENCE   2 extracted excerpts
   ↓
MODEL RUN  4 runs, methodology 1.0.1
   ↓
SCORE      -7  (Mixed or uncertain)
   ↓
CONFIDENCE 65%