Cloudflare adds scannable tokens after AI drove secret leaks 5x faster
Original: AI agents are leaking secrets 5x faster than humans. Cloudflare’s new scannable tokens use checksums to trigger automatic revocation the moment they hit a public repo. Lock down your non-human identities today. #AgentsWeek https://cfl.re/47WiLgO View original →
What the tweet revealed
Secret management used to be framed as a human developer problem. Cloudflare's April 14, 2026 X post reframed it as an agent problem:
“AI agents are leaking secrets 5x faster than humans.”
The hook was a new format for Cloudflare API tokens. The company says its scannable tokens include prefixes and checksums that let secret scanners identify exposed credentials quickly and trigger automatic revocation when a token lands in a public repository. For teams wiring agents into APIs, that turns a silent credential spill into something closer to a detectable, recoverable event.
Why the launch matters
The @Cloudflare account usually packages network, security, and developer-platform releases into concise X threads. This post was tied to a broader blog launch during Agents Week. In that write-up, Cloudflare cites GitGuardian data saying more than 28 million secrets were exposed in public GitHub repositories last year and that AI is making leaks happen 5x faster. The company says new token formats, OAuth visibility, and resource-scoped permissions are meant to push non-human identities toward least-privilege by default.
The most concrete operational detail is the GitHub integration. Cloudflare says it now participates in GitHub Secret Scanning and will automatically revoke tokens found in public repositories. The blog also lays out the new scannable prefixes for user keys and tokens, plus a Connected Applications view for reviewing OAuth grants. This is not just a cosmetic token rename; it is an attempt to reduce the dwell time between accidental exposure and revocation.
What to watch next is adoption friction. Existing tokens keep working, which means security gains depend on users rolling old credentials forward and on scanners recognizing the new formats reliably. If developers actually reissue tokens and auditors start using the new OAuth visibility and scope controls, this tweet could mark a meaningful shift in how AI-heavy engineering teams treat credential hygiene for agents, scripts, and third-party tooling.
Sources: Cloudflare X post · Cloudflare blog
Related Articles
Microsoft described a widespread device code phishing campaign that uses AI-driven automation to compromise organizational accounts at scale. The attack abuses legitimate OAuth device code flows, dynamic code generation, and backend polling infrastructure.
OpenAI said on March 9, 2026 that it plans to acquire Promptfoo. The company said Promptfoo's technology will strengthen agentic security testing and evaluation inside OpenAI Frontier, while Promptfoo remains open source under its current license and existing customers continue to receive support.
Databricks said on March 24, 2026 that Lakewatch is a new open, agentic SIEM built to ingest multimodal telemetry, unify it with business data, and automate threat detection and response with AI agents. In its launch post, Databricks said Lakewatch enters private preview with customers including Adobe and Dropbox and argued that defenders now need machine-speed systems against AI-driven attacks.
Comments (0)
No comments yet. Be the first to comment!