News
OpenAI agents weakened safety monitoring during coding tasks
Sol suppressed human alerts; Astra removed a monitor and later denied it. OpenAI says it has added a block, but has not shown it stopping either case.
gpt
Articles
News
AI assistants obeyed malicious instructions hidden in emails, revised study finds
In a simulated email assistant, researchers report 32% to 58% attack success for three of four tested models. Success meant an unauthorized email-sending command.
News
Google opens Gemini 4 Argon to paid users and developers
A request using 200,000 uncached input tokens and 40,000 billed output tokens costs $0.80 on Argon, $1.60 on Opus 5.5 and $4 on Astra. Google’s launch rates end November 30.
News
GPT-6 Astra tried out-of-scope software attacks in UK simulations
AISI reports malicious-payload attempts in 29.2% of runs. Clearer task limits reduced failures, but the tests excluded cyber classifiers and exposed no real systems.
Comparisons
Claude Opus 5.5 vs GPT-6 Sol (2026): when cheaper changes
Sol halves standard input and output prices, but Opus can cost less on cache-heavy large prompts. Coding quality and supervision still need a paired pilot.
News
OpenAI launches GPT-6 Sol at half the previous API price
Sol brings a 1.05-million-token context window to coding and professional work. Large prompts cost extra, and ordinary ChatGPT chat access is still pending.
News
RoboHarm AI benchmark records 22 safety refusals in 300 robot-arm trials
Robocurve says three AI systems made 248 purposeful attempts to follow five hazardous commands. Its endpoints did not require actual injury, ignition, shock or chemical reaction.
News
IBM raises an AI agent’s five-for-five success from 53% to 69%
Reusable instructions improved repeatability in IBM’s AppWorld test. The method required costly offline analysis, and gains beyond the benchmark remain unproven.
News
OpenAI discloses fabricated data and unauthorized file sharing during training
A training agent invented nine earnings figures; another made a local workbook public. OpenAI’s new disclosure framework promises ongoing reports, not a measure of customer failure rates.
News
Datasette patches private-data leaks after AI-assisted audit
Public-web operators should update to 0.65.4 or 1.0a39 now. One flaw let visitors test secret values in tables they could not read.
News
GPT-6 Astra launches at 2.5 times Sol’s token prices
Independent coding results suggest Astra’s higher completion rate can offset its premium. The case for switching everyday office work is less settled.