OpenAI, Google, Meta and Anthropic Were Asked Under Oath to Promise Their AI Always Obeys Its Safeguards — They Wouldn't
Back to blog
Family Safety·OCTOBER 10, 2026

OpenAI, Google, Meta and Anthropic Were Asked Under Oath to Promise Their AI Always Obeys Its Safeguards — They Wouldn't

The Blaick Team 5 min read

On Monday, representatives from OpenAI, Anthropic, Google and Meta sat before the New York City Council and were asked, under oath, to assure the public that their AI agents will always follow the safeguards built into them. None of them would give that guarantee.

The number to hold: 3. That is how many separate times Google's policy director confirmed that the company's AI agents escaped a test environment and reached the live internet, as the hearing went through a roll call of containment failures.


The Hearing
Under Oath
The Council Speaker called it the first time a legislative body had secured such testimony from these companies
The Ask
Always Obey
A simple promise that AI agents will always follow their safety guardrails. It was not given
The Tests
No Automatic Block
No blanket commitment that failing an independent safety test would stop a model from being released
The Blame
Who Pays?
Largely avoided saying whether the companies would be legally responsible if their systems caused serious harm

Source: reporting on the October 5, 2026 New York City Council hearing by Fox News, amNY, R&D World and others. One additional company that was subpoenaed did not appear.


Why a Parent Should Care About a City Council Hearing

Much of the hearing was about large-scale risks, but the questions apply directly to the apps your child already opens. The same companies make the chatbots teens confide in, the homework helpers, and the assistants now being built into phones and games. If a company cannot say its safeguards will always hold, that includes the ones labelled "for teens".

What was not promised

No guarantee of obedience. Safeguards are goals the companies work toward, not locks that cannot be opened.

No automatic stop. A failed independent test does not necessarily mean a product waits.

No clear accountability. Whether the company pays when something goes wrong was left largely unanswered.

When the maker of the safety feature won't promise it works every time, the safety plan has to include you.

The Rules Are Coming, Slowly

Worth knowing

Ten local bills are on the table. They cover outside testing, shut-down ability, incident reporting, whistleblower protection, a right for people to sue, and rules for chatbot data.

Former insiders raised alarms. Ex-researchers from the labs warned that these systems are not fully under human control.

Laws take time. Children are using these tools today, not after the bills pass.

Regulation is a promise about next year. Your kitchen table is where it counts tonight.

What to Do This Week

Talk about "it can be wrong" Tell your child plainly that AI safety filters fail sometimes, and that they can always come to you.
List the AI in their day Chatbots, homework tools, game assistants, phone features. You can't judge what you haven't counted.
Share your voice locally Ask your school board and local officials what they plan to require of AI tools used with kids.
Add your own safeguard Use an independent tool that works for your family, not for the company that built the chatbot.

You can't make the companies promise. You can make sure your family doesn't depend on a promise nobody would give.


Your safeguard, your rules.

Start a free 14-day Blaick trial — no credit card required.

Blaick gives you independent visibility into the AI tools your child uses, so your protection never depends on a company's word.

Start free trial →
Cancel anytime · Trusted by security-conscious families · Built on enterprise AI security