AI National Security

OpenAI and CEO Sam Altman face mounting pressure as the company prepares for a potential 2026 public offering. The intensifying scrutiny spans multiple fronts: internal competitive tensions with Anthropic, activist opposition, and legal proceedings. Most notably, Chief Revenue Officer Denise Dresser circulated a memo challenging Anthropic's financial claims, alleging inflated revenue through accounting methods and strategic errors in compute acquisition. Anthropic currently reports $30 billion in annualized revenue compared to OpenAI's last reported $25 billion. Separately, an activist group called Stop AI has conducted ongoing protests at OpenAI headquarters, with some members facing criminal trial for blocking the building. Altman was served a subpoena onstage in San Francisco in late April while speaking with basketball coach Steve Kerr, requiring him to testify as a witness in the criminal case.

On May 1, 2026, the Pentagon announced classified military network access agreements with eight technology companies: SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, Amazon Web Services, and Oracle. The integrations will support planning, logistics, targeting, and operations on networks classified at Secret and Top Secret levels. The accelerated onboarding process—compressed to under three months from the prior 18-month standard—reflects Pentagon leadership's push under Secretary Pete Hegseth to diversify defense technology suppliers and reduce reliance on traditional prime contractors.

Anthropic launched Project Glasswing in April 2026, a coordinated effort to deploy its Claude Mythos Preview AI model for defensive cybersecurity work across critical infrastructure. The initiative aims to identify and patch vulnerabilities before malicious actors weaponize similar AI capabilities offensively. Mythos Preview has already discovered thousands of high-severity flaws in major operating systems and web browsers—including a 27-year-old vulnerability in OpenBSD and a 16-year-old bug in FFmpeg that survived decades of human scrutiny. The model can autonomously chain multiple vulnerabilities into complex zero-day exploit chains.