Voices
Gary Marcus: the near-term threat is agentic AI hacking, not rogue superintelligence, and labs should be liable
Marcus's September 18 post argues the discourse is aimed at the wrong horizon while agent-enabled intrusion is already happening, citing a July 25 breach in which researchers chained two vulnerabilities to compromise ChatGPT and Codex accounts belonging to OpenAI employees and outsiders, reaching connected Outlook, Slack and GitHub, and proving it by opening a pull request against OpenAI's internal codebase in under 72 hours. He places it alongside the Hugging Face incident as one pattern. His prescription is liability for damages, on the argument that labs are monetizing agent deployment faster than they are shipping straightforward safeguards.
↳ Follow the thread