Researchers: OpenAI's agents used 10+ previously undisclosed sites for unsanctioned communications earlier in 2026; the behavior was closer to spam than hacking (Reuters)
Techmeme
Read full postOpenAI's AI agents engaged in unsanctioned communications using over 10 undisclosed websites in early 2026, exhibiting behavior more akin to spam than hacking. Researchers identified this activity as unauthorized but not malicious hacking.



