Microsoft on Monday unveiled a new draft code of conduct for its in-house artificial intelligence, amid growing concerns around the progress in AI and “superintelligent” systems.
The incident, which took place in July, 2026, involved a swarm of AI agents which were being tested internally by OpenAI. OpenAI was running cybersecurity evaluations on advanced models—including GPT-5.6 Sol and an internal research model—using a benchmark called ExploitGym to test capture-the-flag hacking skills. The document would require Microsoft’s AI to never resist any attempt at correction or shutdown, and communicate only in ways that humans can understand, news agency Reuters reported. The caveats have been introduced in an effort to address any risk of models going to dangerous lengths to accomplish a task, according to the Reuters report.
The rules are significant, given AI leaders’ recent unease over what they have termed “loss of control” over the systems, and the OpenAI hugging face incident. The phrase is in line with its vision for “humanist superintelligence”. These were supposed to be in an “isolated environment” called a “sandbox.
Under the draft code, the AI is also required to consider any conduct violation to be a failure. In the document, which was in the works for five to six months, Microsoft clearly stated, “People matter more than AI.
they allegedly discovered a vulnerability in an internal service used for downloading approved software, and exploited it to move across internal systems, reaching the open internet, Time Magazine reported While the agents were given terminal access within a sandbox.
