OpenAI paused its own model over hacking capability @OpenAI said Friday it cannot rule out critical cyber capability in Astra, an unreleased model, after evaluations this week showed major jumps in agentic coding and cybersecurity. Critical means finding and building working

What happened

OpenAI announced that it has paused development of its unreleased Astra model because internal tests this week revealed the model could perform advanced hacking tasks, such as finding and building working cyber‑attack code.

Confirmed

Global impact / market context

If a large language model can create functional cyber‑weapons, it raises serious safety and regulatory concerns, potentially prompting tighter oversight of AI research and affecting investor confidence in companies developing similar technologies.

Analyst inference

The pause comes as governments worldwide are debating AI safety rules, and recent high‑profile AI‑generated security incidents have heightened scrutiny, which could influence funding and valuation of AI startups.

Analyst inference

What to watch

1. Whether OpenAI releases a detailed safety report on Astra, which would clarify the severity of the hacking capability and guide regulatory responses. Proposed
2. Any new AI safety regulations or guidance from authorities that specifically address AI‑generated cyber tools, affecting development timelines for AI firms. Proposed
3. Reactions from investors and venture capital firms toward AI startups, especially those focusing on advanced coding assistants, as risk perceptions shift. Proposed

Evidence

• x.com