Cloudflare announced it has automated triage of its bug bounty program using Anthropic’s Claude Sonnet model for about $58 per month, a fraction of the $200,000 monthly cost estimated for its proprietary Mythos security model. The chief security officer, Grant Bourzikas, said the AI system reduces manual effort and improves efficiency.

The company previously processed all bug reports manually. By replacing many third‑party security tools with internally built applications—some developed with AI assistance—it has cut costs and altered its security stack.

Bourzikas cautioned that Cloudflare’s specific circumstances, including its large internal security engineering team and experience creating over 200 autonomous agents, make its build‑versus‑buy decision atypical for other organizations. He emphasized that companies should not assume they can replicate the approach without comparable expertise.

Stephanie Cohen, Cloudflare’s chief strategy officer, said AI will reshape how vendors interact with customers, shifting from selling packaged software to providing forward‑deployed engineers who continuously maintain solutions. She linked recent reductions of about 1,100 jobs to AI‑driven automation, suggesting overall headcount may return to prior levels.

Cohen noted that AI companies currently rely on subscription revenue while struggling to monetize the data they scrape for large language models. Cloudflare plans to offer a content‑access intermediary service that would let AI firms pay publishers for content, potentially using micropayments, and charge for the service.

The company also faces scrutiny over its role as an intermediary, with critics questioning whether publishers would trust a platform that could later alter terms. Cohen defended the strategy, citing Cloudflare’s decision to provide free SSL connections to all customers as evidence of its commitment to a better internet.