AI · Sep 2, 2026
Researchers fear safety disaster ahead of OpenAI’s Astra releaseOpenAI faces 30 more lawsuits tied to Tumbler Ridge shooting
Analysis of reporting by TechCrunch. How we work

Edelson PC added 30 Tumbler Ridge complaints and, for the first time, an aiding-and-abetting count. Intent is the new element. The revolving-door account is the old fact that still matters most.
Why it matters
Edelson PC, which filed seven Tumbler Ridge family suits in April, filed 30 more this week for teachers, a principal, and students who were in the building and not physically shot. TechCrunch reports the new complaints are the first to accuse OpenAI of aiding and abetting the shooting rather than only failing to prevent it.
That upgrade is deliberate. Negligence is about a missed warning. Aiding and abetting is about help with intent. The firm is trying to put OpenAI's referral decision, account deactivation, and command chain in front of a California federal judge as a single story of assistance.
From the desk
TechCrunch is careful: the aiding claim 'is likely to face early dismissal challenges' because intent is hard. The complaints name Chris Lehane as the official who told staff not to call Canadian police, then admit the evidence is 'on information and belief.' Kwon's on-record denial is unusually specific: Lehane was not involved in the original referral, and investigators do not report to him. If that holds in discovery, the Lehane-as-spin-doctor narrative collapses into atmosphere.
What does not collapse as easily is the account story. Staff wanted a tip. Leadership said the logs were not 'imminent and credible.' The company deactivated the user. She made another account. The new complaints also throw OpenAI's November 2025 office lockdown at that imminence standard: the company called San Francisco police and circulated a name and photo while saying there was no indication of an active threat. That contrast will survive even if Lehane walks off the page.
OpenAI is already living in a cluster — FSU, suicide suits, and the Hugging Face sandbox escape during cyber evals. Adding 30 workplace-adjacent plaintiffs is how you keep the cluster in the news. Signal Desk's read: watch the ban-and-reactivate mechanics, not the celebrity of the global-affairs officer. If OpenAI cannot show a hard identity ban, every future referral decision looks like theater.
Context
Jesse Van Rootselaar's February 10 attack in British Columbia killed eight people including family members and six at the school. Kwon told TechCrunch the referral judgment is 'not infallible' but 'always rooted in looking out for this balance for people.' Edelson said they are not laying out all evidence yet and that Lehane and Altman will be critical witnesses.
Who feels it
- OpenAI legal
- Two theories now: negligence from April, aiding from this week. Budget for an early dismissal war on intent.
- Other labs
- Write down the override path for law-enforcement tips. 'Deactivate' that is not a ban is now exhibit language.
- Insurers
- Aiding-and-abetting language is how plaintiffs try to punch through product-liability comfort. Expect exclusion fights.
What to watch
- Whether the court consolidates the April and September complaints.
- Documented proof, or not, that a second email restored full ChatGPT access after deactivation.
- Any change to OpenAI's public imminence policy after the SFPD comparison.
Companies: OpenAI
More on OpenAI
AI · Sep 22, 2026
Opus 5.5 hands risky cyber prompts to an older Claude model, and that routing is the real launch storyAI · Sep 3, 2026
One bad morning for ChatGPT, Claude, Grok and Gemini shows how fragile everyday AI has becomeAI · Sep 19, 2026
The AI regulation smackdown isn’t overAI · Sep 16, 2026
A brief history of AI executives calling for regulation