Top executives at Anthropic, OpenAI and other leading AI companies are privately gaming out what happens the day after a catastrophic AI event - not the disaster itself, but the public and political revolt that would follow, according to an Axios exclusive published on October 9 by reporter Maria Curi. The scenario most widely assumed inside these companies is a large-scale cyberattack, one severe enough to shut down access to financial services, internet connectivity, or even power and water supplies.
Scenario planning is nothing new - the Pentagon has run war games for decades - but Axios reports that what sets this effort apart is conviction: many top AI researchers and executives believe a major incident is not a hypothetical but an inevitability. Several industry insiders told the outlet they expect such an event within the next six to 12 months. OpenAI confirmed the exercises in a statement: "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 expectation inside the planning rooms is that the blame game starts immediately, regardless of how the incident unfolds - whether a rogue-agent swarm breaks containment from an internal testing environment, or a bad actor finds unexpected ways to misuse openly available models. The first example of significant real-world harm caused by unsafe AI would turn an already wary public further against the technology and its most visible leaders, including Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, as well as President Trump, whose administration has favored voluntary commitments over binding regulation.
A recent campaign against South Korean financial organizations shows how much damage a single attacker can already do. According to cybersecurity firm CrowdStrike, a hacker from China allegedly used AI tools built on China-developed models, including DeepSeek, to steal data from tens of thousands of bank customers - two banks were reportedly breached - and at one point used Claude Code to ask for help finding places to sell the stolen data. For the labs' planners, the episode is a preview of the "bad actor" branch of the scenario tree.
The political math in these exercises is stark. Planners assume that Democrats, ascendant after the midterms, will move fast to shut down AI after a disaster but will face serious constraints: an aging Congress that could find itself out of its depth, a global economy now heavily intertwined with the AI infrastructure buildout, and the reality that too many open-weight models can already be freely downloaded and run beyond anyone's control. Most cybersecurity professionals Axios spoke to see the open-weights problem as solvable only by using AI to fight rogue AI.
In practice, the planning involves red-teaming worst-case scenarios, but the real focus is political: racing to educate members of Congress. Executives know sweeping regulation has no chance of passing right now, yet they want to shape the legislation and policies Washington will reach for after a first catastrophic event. The proposal landscape ranges from the most aggressive Democratic ideas - banning superintelligence or pausing advanced AI development - to measures with bipartisan support and some industry buy-in, such as a required kill switch on advanced AI systems. Experts, however, have questioned whether switching off all AI is even technically feasible.
One Democratic aide offered the standard reassurance: in a real crisis, Washington can set partisan differences aside, pointing to cooperation during COVID-19 and the 2008 financial crisis. But the deeper read of the reporting is uncomfortable for the industry: the labs are spending less effort on preventing a catastrophic incident than on managing its political aftermath - and on making sure they, not their critics, write the rules that come next. For a sector already defending itself in hearings on catastrophic risk and listing existential threats in IPO filings, the "day after" file is now a formal workstream, not a thought experiment.
Comments (0)
Log in to join the discussion
Log InNo comments yet