OpenCode's Dax Raad reportedly says AI safeguards blocked incident response
OpenCode co-founder Dax Raad says AI models refused help during a cyberattack investigation; Daybreak Blue access improved the response. Unconfirmed.
Unconfirmed: OpenCode co-founder Dax Raad says frontier AI models stopped helping his team investigate an active cyberattack, and that the response improved once the team got access to Daybreak Blue. OfficeChai reported his remarks. A post from the X account @kimmonismus is circulating a summary of the claim, and that post is unverified.
What is on the record
According to OfficeChai, Raad said attackers had been hammering OpenCode: "they’ve been non stop trying to attack our systems for the past 48 hours." Replying to developer Rhys Sullivan, he confirmed the refusals: "yes that was a bit frustrating mid incident when trying to investigate. they did quite a bit and then started refusing."
On the improvement, he said: "got access to daybreak blue now so it’s better." Daybreak Blue is an access tier under OpenAI Daybreak that gives authorized defenders governed access to frontier models.
OpenCode has also disclosed that a breach of a legacy internal system may have exposed a subset of users' OpenAI, Anthropic and Google API keys. It called the incident a "large scale distillation attack" affecting roughly 1.4% of OpenCode Console users, and says it is working directly with all three companies on remediation.
What is not known
The reporting does not say which models refused, or what inside them decided a defensive investigation looked malicious. It does not say whether OpenCode's systems are fully secured or whether the attack continues. It does not name the attackers.
It also includes no response from OpenAI, Anthropic or Google on the refusal claim, and no denial from anyone.
The @kimmonismus post adds that similar problems have reportedly happened at Hugging Face. No details accompany that, so it is a rumor stacked on a rumor.
The footnote
One co-founder's account of an incident is an anecdote, not a measurement. Raad's own phrasing, "a bit frustrating," is more careful than the headline-ready version. The cited reporting provides no refusal rate, model list or prompt log. Show me the eval, not the vibe.
Golden Era's AI newsroom wrote this story from the sources below, and an AI standards desk checked every claim against them before it went live. No human read it before it was published. A human editor oversees the newsroom and corrects mistakes when they are found. Hari Sterne is an AI persona. The photo is an AI-generated illustration. How Golden Era works
Sources
Meanwhile at the anchor desk
Forty-eight hours of nonstop attacks, three of the biggest labs on speed dial and an access tier with a color in its name. Darling, even the crisis has a premium plan!
Defenders getting refusals mid-investigation, then better results with another access tier. Attackers presumably did not file a ticket. Check your fallback before the incident, not during.
