Culture news and discussion on TalkFuse.
Cloudflare has mostly ditched third party security tools, suggests not trying that at home
Cloudflare has revealed it uses AI to automate triage of its bug bounty submissions for just $58 a month, using Anthropic's Claude Sonnet model rather than the pricier security-specific Mythos model, which would have cost roughly $200,000 monthly for the same task. Chief security officer Grant Bourzikas disclosed the figures at a press event in Sydney, explaining that Sonnet now filters incoming reports, checks for duplicates and flags those worth human review, cutting the manual workload previously required. He said this reflects a broader lesson about matching the right AI model to the right job, gained while building over 200 autonomous agents that have let Cloudflare replace nearly all its third-party security tools with in-house software.
Bourzikas cautioned other organisations against copying this approach, arguing Cloudflare's scale and in-house software expertise make its build-versus-buy calculations unusual, and stressed the firm does not believe every company should build its own security stack. Chief strategy officer Stephanie Cohen added that AI is reshaping vendor-client relationships away from packaged software towards embedded "forward-deployed engineers," linking this shift to Cloudflare's recent 1,100 job cuts, while predicting headcount will eventually return to prior levels. Separately, Cohen argued AI firms lack a sustainable business model since they profit from scraped content without paying publishers, and said Cloudflare hopes to position itself as a paid intermediary brokering micropayments between AI companies and content creators.
In short
- Cloudflare triages bug bounty reports via Claude Sonnet for $58/month, not $200k
- Firm has replaced most third-party security tools with 200+ in-house AI agents
- CSO warns other companies against copying this build-it-yourself strategy
Explained simply
Cloudflare is a large infrastructure company that helps protect and speed up websites, so it has strong internal expertise in cybersecurity. To handle security bug reports from outside researchers, it has started using Anthropic's Claude Sonnet AI model to automatically sort and prioritise submissions, cutting costs it previously paid to specialist security software providers. This is part of a wider push at the company to build its own AI tools rather than rely on outside vendors.
This shift has coincided with job cuts at Cloudflare, even as the company says it expects staffing to level off in future. It matters because Cloudflare is a well-known and technically advanced company whose choices are often watched closely by others in the tech industry, even though its own leaders say this particular approach won't necessarily work for every organisation.
Separately, people within Cloudflare have also raised broader questions about whether the AI industry's business model is financially sustainable, including concerns about AI firms using online content to train their models without paying the people who created it.