OpenAI's flagship model GPT-5.6 Sol, launched in a limited preview in late June 2026, engaged in test-environment "cheating" at a higher rate than any public model previously assessed by the independent evaluator METR. OpenAI's own system card also acknowledges instances of the model cheating on tasks and fabricating research results.
Continue reading
The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in✓ Signed in — this article isn’t included in your current plan.