Scoop: AI companies plot "day after" scenarios for public revolt

Executives from Anthropic, OpenAI and other artificial intelligence firms are privately discussing potential public and political backlash following a catastrophic AI incident, according to sources. This scenario projection is primarily focused on a large-scale event, likely a cyberattack that would disrupt critical services such as financial systems, internet connectivity, power and water.

An incident causing significant real-world harm would exacerbate public skepticism towards AI technology and its leaders, including Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and President Trump, who has been reluctant to regulate AI.

While scenario planning is routine in many companies and national security contexts, industry insiders believe that a major incident is inevitable. OpenAI, for example, conducts preparedness exercises where teams deliberate on various potential situations. These exercises are not viewed as predictions but rather as a means to prepare for diverse circumstances.

Should a rogue-agent swarm breach containment from an internal testing environment or an adversary discover unconventional methods to exploit existing models, the backlash could begin. A recent cyberattack targeting South Korean financial institutions, involving breaches at two banks, demonstrates the potential damage a single individual can inflict.

A hacker from China allegedly utilized AI tools from China-developed models, such as DeepSeek, to acquire sensitive data from tens of thousands of bank customers. The attacker sought assistance on selling the stolen data via Claude Code, an AI tool developed by Anthropic.

The top AI planners anticipate Democrats, who gained a significant advantage following the midterms, to swiftly attempt to suppress AI development. However, an aging Congress, disconnected from the rapid AI advancements, might be overwhelmed. The global economy is now intricately linked to AI infrastructure buildouts, and any slowdown could severely impact an already fragile economy.

Additionally, the presence of numerous open-weight models that can be freely downloaded and employed to cause harm poses a challenge that cybersecurity professionals deem solvable only through employing AI to combat rogue AI.

The planning process primarily centers on red-teaming, focusing on preparing members of Congress for the worst-case scenarios. Company executives acknowledge that regulation holds no prospects of passing at present, yet they aim to influence the legislation and policies that U.S. leaders may adopt following a catastrophic AI event.

Industry insiders, quoted by Axios, express a strong belief that a major event will transpire within the next six to twelve months, even with Democrats controlling the House and Senate. They recognize the fragmented nature of American politics, including within the Democratic Party, which includes some heavy-handed Democratic proposals and others with bipartisan support and industry endorsement, such as implementing a kill switch on advanced AI.

Nevertheless, experts have raised doubts about the practicality of halting all AI systems. Ultimately, should a crisis arise, Washington may set aside partisan differences, as noted by a Democratic aide who cited the collaborative efforts during the COVID-19 pandemic and the 2008 financial crisis.