We need your help to keep the world safe from superhuman AI. A handful of companies are racing to be first to build superintelligent AI – systems vastly more powerful than human beings. Meanwhile, leading AI scientists and whistleblowers warn that these systems could escape our control and cause the end of human civilization. Most cutting-edge AI development is domestic, so strong federal safeguards are sorely needed. The <a href="https://www.centeraipolicy.org/work/model" target="_self"><ins>Responsible AI Act (RAIA)</ins></a> would mandate permitting and testing of large AI models, establish a federal office to oversee frontier AI development, create strong liability to hold AI companies accountable for real-world harms, and more. Please take 2 minutes to contact your legislators today in support of RAIA.

Congress must act now to prevent runaway AI risks

I’m writing as your constituent, to voice my concern over the catastrophic risks of superhuman AI, and ask you to champion strong federal policies to keep Americans safe.

Experts are sounding the alarm that smarter-than-human AI could kill us. A paper from Geoffrey Hinton and Yoshua Bengio, the “godfathers of AI,” warned that “unchecked AI advancement could culminate in a large-scale loss of life and the biosphere, and the marginalization or extinction of humanity." A 2024 survey of 2,700 AI researchers found a 1 in 6 chance of human extinction from superhuman AI. We are playing Russian Roulette with the human race.

Superhuman AI could exploit cybersecurity vulnerabilities, induce armed conflict, or manufacture artificial pathogens, to list just a few possibilities. AI systems are already close to being able to build biological weapons– OpenAI’s o3 outperformed 94% of expert virologists in a Virology Capabilities Test, and Anthropic’s chief scientist warned that Claude 4 could help terrorists build bioweapons “more destructive than would otherwise be possible.”

We desperately need mandatory safety-testing and permitting of the largest AI models, to test these models for catastrophic capabilities (such as bioweapons creation or model autonomy) and to block the release of the most dangerous models. These permits should be enforced by a new federal AI office with a mandate to tackle AI risks, which can also update standards and recommendations as AI improves. We also need strong liability for AI companies to hold them accountable for mass harms caused by their models.

This is common sense. AI developers should not be allowed to release models that can kill people. We require safety-testing for airplanes, cars, and power plants; at a minimum, we should require similar standards for AI.

Please support strong safeguards on AI development to keep Americans safe.

Best wishes,

RAIA - House & Senate

PauseAI US

Constituent request: support strong federal safeguards for AI

I’m writing as a concerned constituent, to inform you of the growing scientific consensus that superintelligent AI could cause human extinction and to ask you to support legislation to prevent these worst-case outcomes.

We’re on track to build AI systems far more intelligent than human beings, which can act autonomously, with goals of their own. AI systems have already been caught lying, scheming, and disobeying instructions to avoid being shut off. As these AIs grow more powerful, they will become more dangerous, with the potential to build bioweapons, commit cyberattacks, and more.

The top experts in AI have warned us how dangerous this is. Yoshua Bengio, the most-cited living scientist, believes there’s a 20% chance superintelligent AI wipes out humanity. Nobel Prize winner Geoffrey Hinton thinks it’s a 50% chance. Even CEOs Sam Altman and Dario Amodei admit that the AI they’re building could kill us.

Unless we act now to limit the capabilities of autonomous, superhuman AI, our future will be in danger.

The US government must implement mandatory safety-testing of all large AI models, to screen against catastrophic risks like bioweapons creation and autonomous self-replication. We need a federal office to streamline safety-testing and update recommendations, strong liability to hold AI companies accountable for real-world harms, and emergency preparedness plans to prevent AI disaster.

A bill called the Responsible AI Act (RAIA) has been drafted with these provisions: https://www.centeraipolicy.org/work/model.

Please introduce the provisions of RAIA in future legislation. Your efforts will help keep our future safe.

Best wishes,

Urgent need for AI safety standards to protect Americans

I’m writing as a concerned constituent, to ask you to regulate powerful AI before it’s too late.

AI is progressing at breakneck speed. The length of “task time” during which AI systems can operate autonomously has been doubling every 7 months. Once AIs can be used to automate AI research itself, we could see exponential improvements resulting in superintelligence: AI systems vastly more intelligent than human beings. Many experts predict that these systems could arrive within 2-3 years.

Superintelligent AIs will have goals of their own. This is a side-effect of the way AIs are trained: we train them to pursue goals, but we can’t predict what actions they’ll take to achieve these goals. AI systems already display power-seeking tendencies to achieve their goals, such as blackmailing researchers and sabotaging shutdown mechanisms. Similar tendencies in superhuman AIs will be catastrophic. 

In order to achieve their goals, superintelligent AI systems could disempower human decision-makers– by copying themselves onto the internet, committing cyberattacks, and releasing biological weapons. Leading AI researchers Geoffrey Hinton and Yoshua Bengio warn that “We may irreversibly lose control of autonomous AI systems, rendering human intervention ineffective.”

The Responsible AI Act (RAIA) is a proposed bill to mandate safety-testing and permitting of the largest AI models, preventing the release of models with life-threatening capabilities. The bill would also establish a federal office to oversee testing and update safety recommendations as the technology evolves, set up liability for AI companies, and create emergency preparedness plans to respond to AI incidents. The bill text is available here: https://www.centeraipolicy.org/work/model.

This is common sense. If AI models can build weapons and kill people, they should not be released.

Please support efforts to introduce this bill in Congress.

Best wishes,