As cited
Copy frozen at (site build).
ai security
Cloudflare OS goes open source with a record of everything its agents read
Cloudflare open sourced Cloudflare OS, an agent platform that tracks every resource accessed by agents and enforces data access controls on downstream consumers. The platform maintains a record of all reads linked to agent outputs, and validates viewer permissions against the underlying data before displaying results. Cloudflare has used the system internally since May with this audit and enforcement model.
Why it matters: Enterprise security teams evaluating AI agent platforms need to understand how Cloudflare OS provides access control and auditability for sensitive data accessed by autonomous agents, reducing risk of unauthorized data exposure through agent outputs.
- Source published
- First seen by Cybersecurity Tracker
Source attribution
Correction
Correction recorded as of .
ai security
Cloudflare OS goes open source with a record of everything its agents read
No summary had been written when this copy was frozen.
- First seen by Cybersecurity Tracker
Source attribution
Correction
Correction recorded as of .
ai security
Cloudflare OS goes open source with a record of everything its agents read
Cloudflare released the source code of Cloudflare OS, an agent platform its staff have used internally since May. The platform logs every resource an agent accesses and ties that log to any output the agent creates. When another user opens that output, the system shows the content only if the user’s permissions allow access to the original data.
Why it matters: Security teams building or using agent-driven workflows should review whether their tools enforce data access provenance, as missing controls could let unauthorized users view sensitive outputs.
- Source published
- First seen by Cybersecurity Tracker
Source attribution
Correction
Correction recorded as of .
ai security
Cloudflare OS goes open source with a record of everything its agents read
Cloudflare released the source code of Cloudflare OS, an agent platform its staff have used internally since May. The platform logs every resource an agent accesses and ties that log to any output the agent creates. When another user opens that output, the system shows the content only if the user’s permissions allow access to the original data.
Why it matters: Security teams building or using agent-driven workflows should review whether their tools enforce data access provenance, as missing controls could let unauthorized users view sensitive outputs.
- Source published
- First seen by Cybersecurity Tracker