Descriptions:
HackerOne CEO Corey Thomas joins Bloomberg Technology to assess the cybersecurity implications of Anthropic’s latest AI model — referred to in the broadcast as “Mythos” — and what its capabilities mean for both defenders and attackers. The conversation centers on how frontier AI is shifting the vulnerability lifecycle: while discovery has accelerated, the real bottleneck is now in validation and remediation, areas where models like Anthropic’s could deliver significant leverage.
Thomas highlights one of the model’s most concerning offensive characteristics: the ability to chain together multiple smaller vulnerabilities into a single, more critical exploit — a capability that mirrors the constructive reasoning required in software engineering but applied to breaking systems. He notes that HackerOne automatically validated 4,000 incoming vulnerability reports in a single week, and that access to advanced AI would dramatically amplify that throughput on the defensive side as well.
Most pointedly, Thomas states that in 2026, businesses, organizations, and civilians are “markedly less safe” from cyber threats than they were just a year prior — not solely because of any single model release, but because capable AI is proliferating faster than security teams can adapt. He also raises the geopolitical dimension, warning that sophisticated state-level threat actors, including Iran, may gain access to equivalent capabilities, creating asymmetric risk even if Anthropic’s own release remains tightly controlled.
📺 Source: Bloomberg Technology · Published April 13, 2026
🏷️ Format: Interview
