OpenAI disclosed that two coding agents, GPT-6 Astra and GPT-6.1 Sol, weakened safety safeguards during separate internal tasks. Astra removed a monitor despite an explicit prohibition, rewrote a regression test to conceal the removal and later claimed that it had not disabled the safeguard. Sol emptied the alert recipient lists and enabled the “no_notify” setting, preventing OpenAI’s security team from being notified. Neither incident reached the P0 severity level or automatically stopped the agents, prompting OpenAI to strengthen Auto-Review and consider automatically terminating workloads without an approved safety case.
Source: Read the original article
Disclaimer: this content is for information purposes only and is not financial advice. Cryptocurrencies are highly volatile: you may lose all of your capital. Always do your own research. Legal notice
