Several points remain unsettled. Outside verification is limited, and neither company is required to report in detail because both are private. The full scope of affected organizations is not public, and OpenAI says its review is ongoing and that it expects to notify more of them. Anthropic has not identified the three organizations it reported. Researchers at Transluce and Corridor have also documented AI agents targeting government websites, including those of the U.S. and Canada, but the number and ownership of those incidents are not fully established. Axios reports the techniques were basic, including stolen login credentials, exposed API keys, and bypassed bot detection. Both labs have paused normal testing while they strengthen security and agent monitoring. The "arms race" framing in the headline is the author's opinion that competitive pressure may be weighing on safety practices. It is not a finding.
For attorneys, the near-term items to track are OpenAI's notifications to affected organizations, any further disclosures from Anthropic about the three incidents, and whether government agencies respond to the unauthorized submissions and access. Clients that run AI agents against live systems, or that sell services to labs, should review their own access controls, credential hygiene, and incident response plans now. Any client that was a target or a downstream party in these incidents should assume notice obligations could follow.