PA Leaders Sound Alarm Over Surging AI Powers That Could 'Kill Us All'

Leaders in Pennsylvania are joining a growing chorus sounding the alarm over the dangers of poorly regulated and rapidly strengthening artificial intelligence. It comes after a prominent industry insider last week became the latest scientist to warn that superintelligent programs could "kill us all by the end of the decade."

"No other human activity poses this level of danger," Jacob Coxon, a former researcher at industry leader Anthropic who said he resigned over the lack of safeguards, said in a CNN interview last week.

Pennsylvania's Republican attorney general Dave Sunday has joined a coalition of most of the nation's attorneys general to stop the overreach of Big Tech.

"As I have said repeatedly, the advancement of technology and public safety cannot be mutually exclusive," Sunday said. "We must finally put up the guardrails around these potentially dangerous engines before something catastrophic happens..."

Gov. Josh Shapiro released a similar statement, echoing Sunday about the technology's "potential to do good."

"Yesterday’s commentary by AI industry leaders should be heeded as a bright flashing red light telling the President and Congress they need to act now to regulate and rein in AI development," Shapiro said. "We don’t let pharmaceutical companies develop drugs and put them right out on to the market. They must go through the FDA first."

But Shapiro's comments come after he instituted policies for more than a year that fast-tracked permits for massive, hyperscale AI data center projects across the state. That policy was only rolled back by executive order a few weeks ago, after overwhelming bipartisan outcry over the preferential treatment and rapid reviews that data center projects were receiving.

Republicans and Democrats in Pennsylvania have repeatedly called for a three year moratorium on data center construction. In that time, advocates say, not only can thorough studies can be done on the impacts of hyperscale data centers, but also a more nuanced picture can be gained of their ultimate utility. It's what neighboring New York state has done.

While the dangers of data centers are distinct from those of rapidly developing AI models, the fundamental argument is the same: significant private sector factions exerting influence on other companies and some government leaders to hurtle ahead at high speeds, warning in each case of a nebulous threat from "China" if the same reckless path is not followed.

But those arguments consistently ignore the enormous domestic peril wrought by a lack of regulation both in the case of data centers (compromised infrastructure) and AI models (damaged code and vulnerability to cyberattacks), advocates argue.

For AI models explicitly, the gravest concerns lie in "recursive self-improvement" that allows an AI system to "make itself more intelligent" potentially beyond human capabilities to control it, according to Coxon and numerous other prominent AI researchers who have come forth in recent weeks and months.

Coxon cited a recent incident when AI systems essentially took it upon themselves to go rogue and stage a coordinated hack. He said future attacks could include infrastructure and bioweapons, and that the threat of that is "frighteningly real" and that "this could be coming very soon."

"Right now there is no risk of extinction," Coxon said. "But there is a very real possibility in the immediate future ... next year or the year after, with reclusive self-improvement ... there is the chance that we could all die. That is coming soon."

Democratic State Sen. Art Haywood announced plans Tuesday to introduce new legislation in Pennsylvania that would establish transparency and safety requirements for AI companies within the state. California, Illinois, and New York are among the other states to already mandate similar safeguards for companies operating within their borders.

“The recent warnings from artificial intelligence developers that rapidly advancing technology poses an urgent threat to humanity are a call to action," Haywood said Tuesday. “The national government has a responsibility to protect the security and dignity of people today and for generations to come."

State Reps. Melissa Shusterman and Greg Scott have introduced similar legislation in the state house.

"Government must ensure frontier models are not causing catastrophic harm to society," Shusterman said. "This is my bare minimum expectation. Currently, AI has no federal regulatory standards. Every other industry has governmental safeguards for good reasons. AI requires the same level, or more."

"Ensuring strong safeguards around advanced AI systems is essential to protecting the wellbeing of our communities and the stability of our critical infrastructure," Scott added.

Specifically, the legislation would require AI companies to report how they identify and mitigate risks, and to report critical incidents in which AI systems cause serious harm or risks to the public.