Former White House cyber official urges formal math-based methods to verify AI software security

A former Biden White House cyber official argues AI security risks are fundamentally a software verification problem solvable through mathematical proof.

A former top White House cybersecurity official is calling for a government-led push to mathematically verify the security of software underpinning frontier AI models — a technique she says is being overlooked in current AI risk debates. Anjana Rajan, who served as assistant national director for technology security at the Office of the National Cyber Director under the Biden administration, told Inside Cybersecurity that the AI risk debate is at its core 'a software understanding problem' — meaning the ability to mathematically reason about complex systems under normal, abnormal, and hostile conditions. Rajan argues the ONCD should take a larger coordinating role across agencies to advance so-called formal methods, which use mathematical proofs to demonstrate that software is free from vulnerabilities. She previously contributed to a 2024 ONCD report recommending this approach. She has since co-founded a software security firm called Atalanta alongside Jonathan Ring, another former ONCD official who served as deputy assistant director for technology security. The call comes alongside a June 22 report from the Soufan Center, a national security think tank, focused on software understanding and its implications for national security. The convergence of the think tank report and Rajan's public advocacy signals a coordinated effort to move formal verification methods higher on the federal policy agenda.

Why it matters

As AI systems become embedded in critical infrastructure and national security applications, unverified software vulnerabilities pose significant risks that current AI safety debates may not adequately address. Formal verification methods, if adopted at scale, would represent a significant shift in how the U.S. government certifies the security of AI systems.

What's next

Whether the current ONCD under the Trump administration picks up or rejects the formal methods framework championed under Biden will be a key indicator of continuity in federal AI security policy.

Key facts

• Anjana Rajan served as assistant national director for technology security at ONCD under the Biden administration
• Rajan co-founded software security firm Atalanta with former ONCD colleague Jonathan Ring
• ONCD published a 2024 report on using formal methods to mathematically prove software security, which Rajan helped produce
• The Soufan Center published a report on software understanding and national security on June 22
• Formal methods are defined as the ability to mathematically reason about complex systems under normal, abnormal, and hostile conditions
• Rajan is calling for ONCD to take a lead coordinating role across agencies on formal verification for AI software

Bias & framing notes

This story comes from a single trade publication based on a direct interview with Rajan, who has a commercial interest in software security through her firm Atalanta. No independent or opposing perspectives are included, and the reporting appears largely to amplify Rajan's own policy advocacy without critical examination.

NewsClear — neutral news & congressional tracking · Bill of the Week