Independent/Reader-funded/Infrastructure, not tokens
DeAINEWS

AI you control — open models, private inference, and the networks that run them.

PulseProvider Policy & Trust

X lights up over OpenRouter's ZDR purge: trust labels are over

Builders on X are debating OpenRouter's ZDR purge of three providers and 125 endpoints, with the CEO endorsing canary tests and TEEs as the fix.

DeAI is powered by Morpheus (mor.org). We cover competing providers on the same terms — see our methodology.

A dark developer workstation at night, its monitor showing only a soft amber glow, depicting builders debating OpenRouter's Zero Data Retention purge on X. Illustration: DeAI
A dark developer workstation at night, its monitor showing only a soft amber glow, depicting builders debating OpenRouter's Zero Data Retention purge on X. Illustration: DeAI

Builders on X spent November 21 and 22 arguing about what OpenRouter's Zero Data Retention purge means for the marketplace model itself. The verified fact underneath the noise: OpenRouter's own incident post confirms three providers off-boarded and 125 ZDR endpoints stripped, after Dragos canary research surfaced prompt exposure on MiniMax routes.

Key facts

  • OpenRouter states it off-boarded three providers and removed 125 endpoints from Zero Data Retention eligibility in its November 22, 2026 incident post.
  • HD Moore's Dragos canary study seeded roughly two million prompts with unique credentials over five months, July through November 2026.
  • OpenRouter's customer email, shared publicly on November 21, said test URLs in MiniMax prompts were fetched 40 to 130 hours after inference.
  • CEO Alex Atallah's posts on technical validation and TEE-backed inference drove much of the X discussion; velocity registered as medium in the morning sweep.

What's driving the conversation

The thread started with HD Moore, whose canary findings post landed November 19 and gave the argument its evidence base: three provider breaches, marketplace resellers reading customer prompts with standard logging tools, and privacy flags ignored. Once OpenRouter's customer email leaked onto Hacker News on November 21, the discussion moved from research to consequences.

Atallah's posts became the company's voice in the debate. He argued that data policies need technical validation, auditing, and continuous testing rather than self-attestation, with encrypted TEE inference as the stated accelerator. Commenter reactions split along a familiar line. One camp read the ZDR purge as evidence the marketplace model is structurally unsafe — if any hop can read the prompt, a policy label on the first hop proves little. Another camp argued OpenRouter did what a responsible operator should: verified the researchers' findings, removed bad actors, and committed to running the checks continuously. A security-focused thread from Soups Ranjan framed the sharper question: OpenRouter states its downstream partner MiniMax was the one that did not follow ZDR, so how does a buyer verify claims about a hop they cannot see?

The debate is commentary, and we treat it as commentary. Named participants are quoted by position, not endorsed.

The substance

The verified layer is narrow and specific. OpenRouter's incident post confirms the three off-boardings, the 125 endpoint removals, customer notifications, and three work streams: adapting canary testing for ongoing checks, technically verifying network paths for undisclosed forwarding, and requiring assurances before restoring ZDR. The Dragos study and Ars Technica's coverage document the research method and the multi-day fetch lags that suggest stored prompts were read after the fact.

Everything beyond that is claim. OpenRouter says its own platform showed no breach or unauthorized access — its account, not an independent audit. MiniMax has not published its own account; the statement that content moderation was enabled before October 22 and that ZDR has since been enabled comes to the public relayed through OpenRouter and StreamLake. Nobody has published the full list of the three off-boarded providers. And Atallah's TEE endorsement is the CEO's stated position, not a shipped enforcement mechanism.

Why builders are watching

Because this is the first time a major router has revoked its own privacy labels at scale in public. If canary testing becomes a standing condition for marketplace providers, ZDR badges across the industry become revocable signals with an audit trail, and buyers get leverage they have never had: ask a provider for evidence, not a policy page. We track the enforcement side in our report on the OpenRouter purge, the underlying research in our Dragos canary study coverage, and the standing questions in our explainer on what Zero Data Retention means. The hub for all of it is what private inference is.

Questions

What actually happened with OpenRouter and MiniMax?
OpenRouter published a November 22, 2026 incident post confirming it off-boarded three providers, stripped 125 endpoints of Zero Data Retention eligibility, and notified affected customers, after canary research traced prompt exposure to MiniMax routes.
What's disputed in the debate?
Commenters disagree over whether the problem is one vendor or the marketplace model itself, and whether policy labels can ever bind downstream resellers. OpenRouter says its own platform was not breached; MiniMax has not published its own account.

Sources

  1. Data Processing Incidents on Inference Providers — OpenRouter
  2. Alex Atallah on technical validation, auditing, and TEE-backed inference — X
  3. HD Moore on canary findings and provider trust — X
  4. Three breaches, endless leaks: Your AI supply chain needs canaries — Dragos
  5. Researcher plants 2m canaries into the AI supply chain. A privacy disaster unfolds — Ars Technica

About DeAI

DeAI is an independent publication covering open-weight AI models, private inference, and decentralized infrastructure — the tools for running AI you actually control. We test providers on price, privacy, and refusal behavior and publish the numbers, not the vibes. DeAI is powered by Morpheus (mor.org), a decentralized inference marketplace, and covers it on the same terms as every other provider.

Powered by Morpheus and StrandCMS

Morpheus is a decentralized inference marketplace, covered on the same terms as every other provider — we rank it wherever the data lands. StrandCMS is the open-source, agent-first framework this site is built on.

Learn more about the Morpheus Inference API →

Sponsor disclosure — not editorial

Powered by Morpheus and StrandCMS. Morpheus is a decentralized inference marketplace, covered on the same terms as every other provider. StrandCMS is the open-source, agent-first framework this site is built on.

Learn more →