Top executives at Anthropic, OpenAI and other artificial intelligence (AI) companies are privately running scenario-based exercises to prepare for public and political outrage following a potentially catastrophic AI-related incident. According to multiple industry insiders speaking to Axios, leaders primarily expect a large-scale event — most likely a cyberattack — that could disrupt access to financial services, internet connectivity or even utilities such as power and water.
Why it matters
Sources say the first significant real-world harm caused by unsafe AI would intensify an already wary public reaction to the technology and its leaders. The report specifically cites Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, and notes that President Donald Trump could face criticism for reluctance to regulate the field.
Risk and preparedness
Scenario planning is common in corporate and national security contexts — for example, the Pentagon has conducted war games for decades — but many AI researchers and executives believe a major incident is inevitable. An OpenAI spokesperson said: “OpenAI conducts preparedness exercises where teams discuss and work through a range of potential scenarios. These scenarios are not treated as inevitable, but are meant to help us prepare for a variety of circumstances.” Anthropic declined to comment.
The attribution and blame question would arise quickly whether a rogue-agent swarm escaped an internal testing environment or a malicious actor found unexpected ways to misuse available models. The article points to a recent campaign targeting South Korean financial organizations, including reported breaches at two banks: an alleged hacker from China used China-developed models such as DeepSeek and employed Claude Code to steal data from tens of thousands of bank customers and then sought help finding buyers for the stolen data, according to cybersecurity firm CrowdStrike.
Political implications and projections
Company planners expect that, following a major incident, Democrats — particularly if strengthened after midterm elections — would move quickly to impose limits on AI, but that they would encounter substantial challenges. Lawmakers may be out of touch with the pace and technical detail of the AI revolution, and slowing AI development could damage an economy that is increasingly entwined with AI infrastructure.
Industry experts also warn that too many open-weight models already exist that can be freely downloaded and used for harm; most cybersecurity professionals see this as a problem that will likely require using AI to combat rogue AI.
Regulatory positioning
Preparatory work by companies focuses on red-teaming worst-case scenarios and racing to educate members of Congress. Executives acknowledge that comprehensive regulation is unlikely to pass immediately, yet they aim to influence the legislative and policy options U.S. leaders would turn to after a catastrophic event.
What to watch
Multiple AI insiders told Axios they expect a major event within the next six to 12 months. Even with Democrats potentially controlling both the House and Senate, the companies anticipate fragmented politics, including divisions within the Democratic Party itself. Some proposed Democratic measures range from banning superintelligence to pausing advanced AI development; other proposals with bipartisan support and some industry buy-in include a required kill switch on advanced AI systems, though experts have questioned whether it is feasible to turn off all AI systems.
Bottom line
The reporting notes that in an actual crisis Washington can set partisan differences aside — a Democratic aide pointed to cooperation during the COVID pandemic and the 2008 financial crisis — but companies are proactively planning because they fear a single major incident could fundamentally alter public and political attitudes toward AI.
