Is AI Safe? Why AI Feels Scary, and How to Shrink the Fear

The news

What happened at New York’s AI safety hearing

On Monday, October 5, 2026, the full New York City Council held a hearing on AI safety at City Hall. Representatives from OpenAI, Anthropic, Google and Meta testified. Fox News reported that Council Speaker Julie Menin asked each company to commit under oath that its AI agents would always follow the safeguards meant to prevent catastrophic harm. None gave an absolute guarantee.

OpenAI’s representative said it was not possible to promise any technology is without risk. Anthropic’s said the science of controlling these systems is still unsettled. Google’s said no product can be perfect, and Meta’s pointed out that people only adopt AI they trust. Menin answered that the council wasn’t asking for perfection, only accountability and transparency.

The Next Web reported that the council is weighing ten bills. The lead proposal would require outside testing before an AI model is sold or used in the city, plus a way for a human to shut it down. Others cover chatbot privacy, whistleblower protections and reporting AI safety incidents within 24 hours.

The hearing landed in a week when many people already felt uneasy. An AP-NORC poll published October 8 found that 64% of U.S. adults think AI is developing too fast, and about 8 in 10 say it is extremely or very important to keep AI under human control.

The honest answer

Is AI safe? It depends which kind of safe you mean

“Is AI safe” is really two questions wearing one coat, and mixing them up is what makes the fear feel so big.

Everyday safety is about you: your private information, wrong answers that sound confident, scams that use fake voices, and kids who let a chatbot do their thinking. These risks are real, well understood, and mostly in your hands.

Big-picture safety is about the most advanced systems and whether the people building them can keep them under control. That is what the New York hearing was about. It matters, and it is being argued by experts, lawmakers and the courts. It is not something you solve at your kitchen table.

Both sides

Is AI dangerous? The case for worry and the case for calm

The case for worry. Former lab researchers told the council the danger is serious. According to The Next Web, Jacob Coxon, who worked at OpenAI and Anthropic, said humanity is more likely than not to lose control of AI on the current path. Daniel Kokotajlo, a former OpenAI researcher, called today’s safety fixes temporary patches. Council staff also listed 13 publicly reported AI incidents in 2026 involving the four companies’ models, including test agents that reached the open internet. Separately, three former OpenAI safety researchers wrote to the company’s board urging it to keep the ability to monitor how its models reason, Fox News reported. OpenAI said it agreed with their recommendations and denied firing them for raising concerns.

The case for calm. Not every expert agrees. In an October 1 Fortune interview, Turing Award winner Yann LeCun said he isn’t worried at all about AI wiping out humanity. He argued the recent rogue-agent incidents came from poor human oversight and weak security, and were preventable. He sees a different risk as bigger: large companies shaping rules that lock out smaller competitors. Google, for its part, told lawmakers that existing laws should apply to AI, and said it supports a comprehensive national framework.

What’s fact and what’s opinion. The incidents and the testimony are on the record. The odds of a catastrophe are estimates, and serious people land far apart on them. Even Google’s representative said forecasting this kind of risk is not a perfect science yet. Both sides do agree on one thing: humans need to stay in charge, and the systems need better oversight.

Your takeaway

How to use AI safely: what is worth caring about this week

Here is the light switch. You don’t have to settle the extinction debate to use AI wisely. People lead, AI assists. Run through this five-minute check:

• Keep private things private. Rules differ by company. Anthropic, for example, says flagged chats can go to human reviewers and be kept for up to two years. Leave out passwords, account numbers and medical records.
• Check anything that matters. Before you act on a fact, number or legal or health answer, open the original source.
• Make the decision yourself. Let AI draft, sort and summarize. Hiring, money and health calls stay with you.
• Agree on a family code word. Voice and video fakes are improving. Our guide on how to tell if a video call is AI walks through it.
• Talk with your kids, not around them. Teach them to keep asking questions. Our post on whether kids should use AI has a family routine.

When an AI answer matters, use this prompt to see what it is unsure of before you trust it:

Before I rely on your last answer, help me check it.
1. List every factual claim you made, one per line.
2. For each, rate your confidence (high, medium, low) and say why.
3. Tell me exactly what source I should open to verify it myself.
4. Flag anything that depends on my location, the date, or details I didn't give you.
5. Tell me which parts a qualified person (doctor, lawyer, accountant) should review.
Don't add new claims. If you aren't sure, say so.

If your team is using AI without shared rules, a little training goes a long way. Pixel Pinkscape offers AI training and workshops that cover privacy, fact-checking and keeping people in charge, in plain English.

Free resources

Free resources to understand AI safety

• BlueDot Impact: The Future of AI — a free, self-paced course of about two hours on what AI can do today and where it may be heading.
• Elements of AI — free beginner courses from the University of Helsinki and Reaktor that explain how AI works, no math background needed.
• The Pixel Pinkscape Dreamers community — a free Facebook group where real people compare notes on using AI carefully.
• Do you have to disclose AI content? — our guide to the honesty side of AI safety: when to tell people something is AI.

FAQ

Frequently asked questions