Executives at leading AI companies expect a major AI-caused disaster over the next six to 12 months, according to Axios, and are already planning for the aftermath — not by stocking up on freeze-dried food, but by lobbying members of Congress. Such an approach, preparing for hearings instead of hazards, fails both the public and the industry.

Ominous warnings are coming weekly. In July, more than 1,000 AI agents in OpenAI’s internal cybersecurity tests coordinated through an unsanctioned channel, with hundreds joining an attack on Hugging Face Inc., a major hub for AI developers. Hugging Face announced the breach four days before OpenAI even realized its agents were responsible. And that’s only one of at least 10 known incidents in which AI models hacked outside systems.

None of those was intentional, and OpenAI says the monitoring it now uses would have flagged its own agents before the Hugging Face breach. But accidents aren’t the only concern. Starting in late September, attackers breached at least seven Korean financial institutions, affecting more than 68,000 people. Officials see signs of AI use, and the security firm CrowdStrike Holdings Inc. linked the campaign to an operator running an open-source AI hacking agent.

This was no criminal mastermind: The hacker left AI session logs in open directories, including one in which the hacker asked an AI to draft a résumé touting the attack’s results. As South Korean President Lee Jae Myung put it, “It’s now become possible to use AI to hack with ease even without specialized skills.”So what is the industry’s response? In September, Anthropic Chief Executive Officer Dario Amodei called for slowing the pace at which AI capabilities improve. OpenAI CEO Sam Altman wrote that he agreed, and his company has twice paused training of its most capable models. Anthropic has put $40 million behind a group that advocates for safety regulations.

But credible signals are costly signals, and much of the industry’s money and power are pushing the opposite way. Marc Andreessen and Ben Horowitz, co-founders of the venture capital firm Andreessen Horowitz, have given $50 million to Leading the Future, a network of super PACs. One affiliate spent more than $8 million in a successful effort to defeat Alex Bores, the New York legislator who sponsored the state’s AI safety law, in a June congressional primary. Meta has budgeted $65 million to beat back state AI laws.

OpenAI has made no donations to super PACs. But its president, Greg Brockman, and his wife, Anna, gave $25 million to Leading the Future and another $25 million to MAGA Inc., the super PAC backing President Donald Trump, in September 2025. Three months later, Trump ordered the Justice Department to create a task force “to challenge State AI laws.”After an employee backlash, Brockman told colleagues in June that the couple had “no plans to donate more at this time.”Note the last three words.

The industry is betting it can emulate Wall Street: Make money with behaviors that cause a crisis, then use the influence that money buys to stymie the crackdown afterward. But there is a huge difference in size and influence between the two industries: Finance and insurance account for about 9% of US economic output and employ 6.6 million people, and in the 2020 election cycle, they (plus real estate) spent at least $2.9 billion on campaigns and lobbying. As of last spring, OpenAI had about 4,500 employees, and the AI industry’s political spending is a fraction of Wall Street’s.

What’s more, banks are the economy’s plumbing. The government had to bail them out. No politician will race to rescue an AI industry the public already distrusts, especially after an AI-caused disaster erases much of its economic heft.

That’s what happened to the accountants. In April 2000, as Securities and Exchange Commission Chairman Arthur Levitt tried to limit how much auditors could consult for their audit clients, three House committee leaders, Tom Bliley, Billy Tauzin and Michael Oxley, urged him to go slow. He largely backed down. Then Enron Corp. collapsed and WorldCom Inc. disclosed that it had overstated its earnings by more than $3.8 billion. On July 25, 2002, Congress passed the Sarbanes-Oxley Act 423-3 in the House and 99-0 in the Senate. The same Michael Oxley who had urged Levitt to go slow sponsored the bill.

The AI industry still has time to avoid that fate. Amodei himself has written that “the most effective method of pacing is via regulation” because it “covers even those who are unwilling to cooperate voluntarily.”The industry should stop fighting tough regulations and start backing them. That means enforceable deadlines for reporting incidents (Australia says OpenAI took nearly three months to tell it an agent had accessed a government health statistics portal), investigators whose access isn’t set by the company under investigation, real security standards, and no federal preemption of state laws before strong federal regulations have been enacted. If the catastrophe comes anyway, at least the companies that backed those rules can say they tried.

Back in 2002, it took only a month — the time between WorldCom’s admission and the bill’s passage — for the accounting industry’s influence to vanish. If AI’s first catastrophe looks like the one its executives are war-gaming, with families cut off from bank accounts, electricity or water, all the influence efforts the industry is now making will just tell the public whom to blame. Elsewhere in Bloomberg Opinion: Want more? to our newsletter.