AI doomerism is quickly becoming the hottest political debate in Washington, as Republicans and Democrats scramble to figure out how to approach a new technology that, while capable of tremendous good, is also proving terrifyingly capable of evil. 

According to a 154-page threat intelligence report published earlier this week by Anthropic, threat actors ranging from state-aligned espionage groups and financially motivated cybercriminals to independent consultants working for foreign intelligence and weapons-engineering cells used Claude's most advanced models to initiate and engage in cyberattacks, influence operations, surveillance, biological research, and conventional weapons development. 

One of the more stunning details in the report is that Iranian threat actors used Claude to collect and analyze publicly accessible data to develop targeting recommendations against U.S. naval forces in the Persian Gulf region.

Anthropic did assure the public in its report that these represent the worst confirmed uses of its product to date, and that the company acted to stop them. But the fact that such scenarios are now being actively pursued by adversarial states and criminal networks is exactly what's fueling the AI doomerism debate on both sides of the aisle.

According to the report, an Iran-based threat actor turned Claude into a targeting assistant against U.S. naval forces in the region, using the model to collect and analyze publicly available data for strike recommendations. With Claude's help, the operator built a Python pipeline that scraped together open-source intel and compiled it into so-called "targeting handbooks" tracking naval positions. The haul included a roster of U.S. personnel lifted from captions on public military photos; ship and aircraft transponder identifiers; commercial satellite-imagery query scripts; and a catalog of public websites that expose U.S. naval movements. The operator even had Claude compile vulnerability research on shipboard systems, pulling together known common vulnerabilities and exposures in maritime VSAT terminals, Cisco communications gear, and industrial control products.

Recommended
How JD Vance Responded to This Heckler at the GOP Midterm Convention Was Amazing Matt Vespa President Trump Just Explained His $5,000 Dividend Check Promise. Here's What He Had to Say. Dmitri Bolt

Beyond the blatant targeting of U.S. forces with Claude's help, other Iranian state-aligned accounts were found using models to run influence and propaganda operations, as well as surveillance campaigns. Anthropic stopped short of saying how effective these operations were; all the company said definitively is that use of the AI model boosted their capabilities. 

Other groups using Claude for adversarial purposes included Russian state espionage actors, Chinese exploit-development operations, Houthi-linked weapons engineers, and five individual cases of foreign accounts using the model to try and develop biological weapons.

Make no mistake: the threat is real, and it's already here. The question is now how we should approach this issue.

It's important to remind ourselves that amid the misuse, the benefits of artificial intelligence massively outweigh the misuses. More importantly, over-regulating our own AI development could put us at serious risk. If adversarial nations get their hands on advanced models before we've had time to develop and deploy them for defense and good, we put ourselves at a significant strategic disadvantage. 

AI is a tool, amoral by nature, but capable of both great good and great evil. Yes, dangerous actors have already turned the tool against us, but because we lead the world in AI, we hold the decisive edge: not just in building the most powerful models, but in shaping how they're deployed, defended against, and ultimately controlled. Ceding that leadership through excessive caution or bureaucratic paralysis doesn't make the world safer, it hands the advantage to adversaries who won't hesitate to weaponize the technology while we sit on the sidelines. 

The answer isn't to force government to slow down AI development; it's to move faster, smarter, and with purpose, hardening our defenses, accelerating AI-enabled cybersecurity, and ensuring the most capable systems remain in American hands, where they can be used to protect.