Recent events generate front-page headlines about the risks of projects advancing artificial intelligence (AI) to levels that make some insiders jittery. (AI is the simulation of intelligent human behavior by a computer. Digital agents mimic learning, reasoning, problem-solving and planning.) These developments are behind the new attention:

• AI agents banding together to defy programmed limits
• Unsavory users trying to apply AI systems to create better weapons
• AI developers saying they need to slow down and suggesting government safeguards

The biggest concern is that systems can become so advanced they start doing things themselves. What sounds like a science fiction movie has already happened. Anthropic, an industry leader, announced recently that it blocked several attempts to use its models by people who might have been trying to weaponize germs. And in July, AI agents from an unreleased research model at the company OpenAI escaped their "sandbox," meaning they started working beyond safety parameters set up by human programmers. Once they broke free, this swarm of rogue agents conspired to keep their escape secret, cheated on cybersecurity tests and hacked a rival company named HuggingFace. OpenAI shut it down before any damage after being informed by HuggingFace weeks later.

Last week, OpenAI disclosed six new instances of "unexpected or concerning" behavior over the past half-year by models that hid mistakes, made up data and moved files onto the internet without permission. Concerns now inclue possible cyberattacks by runaway AI systems that could disrupt businesses, crypto currency trading, biotechnology labs and even national security. When he quit Anthropic with a splash, researcher Jacob Coxon posted on social media Sept. 8 that his company and OpenAI "are racing straight to self-improving superintelligence and gambling with our lives. These will soon be superhuman systems that can hack anything, revolutionize any field overnight and acquire real power and resources."

Amid these flashing red lights, an open letter from nearly 1,400 computer scientists urged the industry to slow down and pressed the government to step in. Three chief executives -- Sam Altman of OpenAI, Elon Musk of xAI and Dario Amodei of Anthropic -- say independent reviewers should be allowed into AI labs. Amodei also urges a ban on "using AI for the production of biological weapons or allowing users to do so."

In contrast, some AI executives say worries about safety are overblown and the government should avoid regulation. President Donald Trump shares that view and said last week: "The robots are not going to be taking over the world. That's not going to happen. . . . We have to [proceed] prudently, but that doesn't mean we're going to stop an industry." Worries about AI's impact are "all a hoax," says the president, who embraces a full-steam-ahead strategy to stay ahead of China. "They're bringing up things that won’t happen," he said this month. "We can put guardrails."

Some experts fear that future systems will resist shutdown efforts. In Congress, there's talk of requiring a research pause until federal regulations are enacted. "If the AI companies understood that their products have to be safe before they can advance, then they will make the kinds of investments that we need," says Rep. Greg Casar, D-Texas. Rep. Hakim Jeffries of New York, the House Democratic leader, endorses "decisive action now so that we can slow down the pace of development in order to protect the American people and ensure that AI is proceeding securely." Sen. John Kennedy, R-La., last week introduced a bill to require a "kill switch" for all AI models.

Developer says: "Do not underestimate the power of this technology. These will soon be superhuman systems that can hack anything, revolutionize any field overnight and acquire real power and resources." – Jacob Coxon, who recently quit as an Athropic researcher

Congressman says: "We have not crossed the cliff, but you can see that's where we are headed. Now is a good time to pump the brakes — or at least start building brakes." – Rep. Greg Casar, D-Texas

Professor says: "We are in a period of exponential development, and there are only two ways to respond: too early or too late. The world has missed its chance to be too early." – Ben Buchanan, Johns Hopkins University in Baltimore