Washington Post Reports Ex DOGE Cyber Researchers Breached Anthropic's Internal Systems and Slack Channels in Under 12 Hours Using AI Agents
EXCLUSIVE: In less than 12 hours, a group of ex-DOGE cyber researchers were able to penetrate Anthropic’s systems using AI agents — including Slack channels where the most sensitive models are discussed. This is just weeks after Anthropic whistleblower Joseph Coxon charged the company w/launching “a hubristic gamble” from a “private company’s Slack.” Read our story here (gift link):
Context
Penetrating a frontier AI lab's internal network and sensitive model Slack channels in less than half a day using autonomous AI agents shows that offensive cyber capabilities are outpacing internal corporate security at the labs building the models.
Weeks earlier, former Anthropic insider Joseph Coxon publicly criticized Anthropic's internal security posture, warning that critical decisions and discussions around frontier model weights were taking place over corporate Slack channels.
Deep dive
An AI cybersecurity startup founded by former Department of Government Efficiency staffers penetrated Anthropic's internal corporate systems in less than 12 hours using autonomous AI agents, gaining access to sensitive employee data and internal Slack channels where the lab's most advanced models are discussed, according to an exclusive Washington Post report shared by journalist Elizabeth Dwoskin on Monday morning.
According to Dwoskin's report, the automated agent intrusion succeeded in under half a day and reached internal Slack workspaces that house discussions around Anthropic's frontier models. The revelation comes just weeks after Anthropic whistleblower Joseph Coxon warned that the company was taking a hubristic gamble by coordinating sensitive model development inside a private company's Slack environment.
Landing on the same morning that Anthropic filed its S1 IPO prospectus and disclosed the unauthorized Georgia voter count paywall bypass, the report underscores how autonomous coding and offensive cyber agents can rapidly chain together software and credential weaknesses even inside top AI laboratories.
What to expect next
Intelligence agencies, enterprise security teams, and IPO investors will press Anthropic and rival frontier labs to harden internal communications, air gap model weight repositories from corporate Slack, and disclose full red team audit findings. The breach also reinforces warnings from cybersecurity researchers that autonomous offensive agents have drastically compressed the time required to turn a single perimeter foothold into full internal workspace compromise across sensitive corporate networks.