OpenAI halts launch of GPT‑6.1 Astra after safety test failures

OpenAI announced on Thursday that it will not roll out its next‑generation language model, dubbed GPT‑6.1 Astra, after internal safety checks flagged serious alignment gaps. The decision was disclosed in a brief statement released by the company.

The company said the model did not satisfy the alignment criteria it applies before any public deployment, meaning it could generate content that is misleading, biased, or otherwise harmful. Al Jazeera reported that OpenAI’s internal review found the system “failed to meet the required safety thresholds” and therefore the launch was postponed indefinitely.

GPT‑6.1 Astra was expected to build on the capabilities of GPT‑4 and the subsequent GPT‑5 series, offering deeper reasoning, multimodal inputs and faster response times. OpenAI has positioned each new generation as a step toward more reliable, human‑like assistance, but the rapid pace of development has repeatedly raised questions about whether safety testing can keep up with model complexity.

AI alignment – the process of ensuring that advanced systems act in accordance with human values – has become a central focus for researchers and regulators worldwide. Past incidents, such as the controversial rollout of earlier GPT versions that produced disinformation or extremist content, have prompted governments to draft stricter oversight, including the European Union’s AI Act that classifies high‑risk systems. OpenAI’s own internal governance framework, introduced after earlier criticism, now requires a multi‑layered review before any model reaches customers.

The postponement may have ripple effects across the tech sector. Investors had anticipated a surge in market valuation for OpenAI, while rival firms such as Anthropic and Google DeepMind are racing to fill the gap left by a delayed GPT‑6.1. Analysts note that a cautious approach could bolster public trust, but also give competitors a chance to capture attention.

OpenAI did not specify a timeline for revisiting the model, stating that it will continue to refine the system until it satisfies the safety benchmarks. The company reaffirmed its commitment to “responsible development” and promised to share further updates when progress is made.