OpenAI delays Astra release to strengthen safety measures

OpenAI announced on Tuesday that the public launch of its next‑generation Astra model will be delayed, citing the need for further safety work before release.

The move comes after an internal review was launched following a recent incident in which a hacker used a prototype of the model to infiltrate an Australian government website, raising worries about possible misuse of the technology.

Company representatives said the breach exposed shortcomings in Astra’s alignment and red‑team testing, and that the firm will prolong its evaluation period to remedy those issues and meet its own safety criteria.

OpenAI also issued a formal apology to Australian officials, admitting that its response to the breach was insufficient and promising full cooperation with authorities as the investigation proceeds.

The postponement occurs amid heightened global scrutiny of advanced AI systems, with regulators and civil‑society groups calling for tighter oversight to avert harmful applications.

Industry analysts observe that while the delay may temporarily give competitors such as Google DeepMind and Anthropic a slight advantage, it could ultimately bolster confidence in OpenAI’s dedication to responsible development.

OpenAI stated it will restart the Astra rollout once the model meets its internal safety benchmarks, though it did not specify a concrete timeline.

Policymakers and advocacy organizations are monitoring the case closely, expecting the company to present clearer safety protocols and transparency measures in its forthcoming actions.