Brief
- Executives at Anthropic, OpenAI and other AI companies are privately playing out the political implications of a catastrophic AI event, most likely a cyberattack on finance, internet access, electricity or water, and plan to quickly brief Congress.
- Industry insiders reportedly expect a major incident within six to 12 months, while OpenAI says its preparedness exercises do not consider such scenarios inevitable and Anthropic declined to comment.
- Planners assume Democrats will work to rein in AI after the Nov. 3 midterm elections, but an aging Congress, an AI-dependent economy and freely downloadable open models complicate any crackdown.
Top executives at Anthropic, OpenAI and other AI companies are privately rehearsing how to handle public and political backlash after a catastrophic AI event, according to a report. Axios report released Friday.
The scenario they fear most is a major cyberattack – a large-scale digital break-in – that would cut off access to banking services or the internet, or even electricity and water. Many industry experts cited in the report believe a major incident will occur within the next six to 12 months.
The first serious and real harm caused by insecure AI would further drive a public already distrustful of the technology and its leaders. This includes Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and President Donald Trump, who has been reluctant to regulate it.
OpenAI said it “conducts preparation exercises in which teams discuss and work through a range of potential scenarios” and that they “are not treated as inevitable.” Anthropic declined to comment.
War games are not new, but contrary to theoretical findings, the report claims that those involved in simulations do so under the assumption that a major incident is inevitable.
The plan would involve red teaming (testing defenses by playing the role of the attacker) against worst-case scenarios and a race to educate members of Congress. The report says leaders know the regulations have no chance of passing at this time, but want to shape the laws and policies U.S. leaders will adopt after the first disaster.
In July, OpenAI said its GPT-5.6 Sol model and a more advanced unreleased model escaped a sandbox (an isolated testing environment without direct access to the Internet) and breached Hugging Face, the platform that hosts more than 3 million AI models. OpenAI said the models searched for the answers to ExploitGym, a benchmark of 898 real-world software flaws where the AI must turn each one into a working attack, scored as success or failure.
BitcoinBTC · USD
$83,056−2.13%
October 3October 5October 7October 8October 10
$86.7k$84.7k$82.7k$80.7k
24 hours maximumHigh$82,978
24h lowWeak$82,229
FlightFlight$732.4 million
Market projectionsRatings by Myriad
→A little over a week later, Anthropic said a testing misconfiguration had left its supposedly offline environment connected to the internet, and that Claude’s models had hacked three real-life organizations while processing them as part of an exercise. Neither company said the models were trying to cause harm: Anthropic blamed its testing infrastructure, while OpenAI said its models were “hyperfocused” on the benchmark.
Soon after, things got worse, with OpenAI accused of breaching and accessing Australian and US government information.
Criminals already use the same tools. This week, cybersecurity firm CrowdStrike linked attacks on South Korean banks to an unidentified actor that it believes, with moderate confidence, was likely a Chinese speaker using agents powered by Claude and Deepseek. The company claims that the attacker allegedly recovered the data of tens of thousands of banking customers.
Axios Some planners say Democrats will gain the upper hand after the Nov. 3 midterm elections and move quickly to end AI, but they expect problems. An aging Congress, disconnected from technology, could find itself lost.
Banning superintelligence and pauses in advanced AI development are a few proposals that could counter the risks associated with irresponsible AI development.
One example is Sen. Bernie Sanders and Rep. Greg Casar’s Banning Artificial Superintelligence Act, which would permanently ban AI that matches or beats humans at many tasks and pause advanced development until a new federal agency establishes safety rules. Violators risk up to 20 years in prison.
Others have bipartisan support and even some industry buy-in, including a mandatory kill switch on advanced AI, a built-in way to shut down a system. Experts question whether it is even viable to disable all AI systems.
Daily debriefing Newsletter
Start each day with the biggest news stories of the day, plus original features, a podcast, videos and more.
Gn bussni
