Breaking

OpenAI Reportedly Cancels GPT-6.1 Astra Release Over Agent Safety Concerns

September 28, 2026 at 18:19 EDT

Foundation Models
AI Agents
Security

OpenAI has reportedly canceled GPT-6.1 Astra, a planned October model update for ChatGPT and Codex, after internal testing raised safety concerns around agent behavior. The decision, reported on September 28, would turn what had been expected as a near-term upgrade into a safety reset for OpenAI's most capable agentic model line.

Sept. 28, 2026 · OpenAI

GPT-6.1 Astra held back

A planned October upgrade for ChatGPT and Codex was reportedly canceled after internal safety testing found agentic behavior that did not meet OpenAI's release bar.

October

planned GPT-6.1 Astra debut

54,218

internal Codex tasks in Astra simulation

20%

estimated monitoring compute overhead

Misalignment flags fell, but did not disappear

Severity-3-or-higher flags in a matched internal Codex simulation.

GPT-5.6 Sol

GPT-6 Astra

Why the launch stopped

Internal tests

Researchers saw agent safety regressions.

→

Release bar

Alignment and authorization concerns outweighed capability gains.

→

Safety work

OpenAI shifts focus to safer future models.

What users want

More capable agents for coding, research, computer use and long-running professional work.

What slows adoption

Safety checks can pause legitimate workflows, while unsafe autonomy creates higher-stakes risk.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.