In the same week as its key AI researcher Jacob Coxon’s headline-making exit from the company, Anthropic released a threat intelligence report. The eye-widening details laid therein shed light on notable cases where its Claude AI models had allegedly been used to develop biological weapons and possibly even carry out a suspect Russia-linked cyber espionage campaign against Ukraine.
The artificial intelligence heavyweight also said it identified an Iran-linked actor using its own American chatbot creation to target US Navy warships. The Dario Amodei-led company’s fear-inducing report surfaced as its own crucial executives have been speaking out against AI safety concerns. Sharing his exit message, Coxon dropped a dire existential AI warning on Tuesday, saying that artificial intelligence was likely to kill us all by the end of the decade.
Back in July, more than a thousand employees of frontier AI companies signed an open letter titled ‘Pacing the Frontier,’ making their case in favour of slowing down advanced AI’s development. Anthropic’s co-founders Dario Amodei, Jack Clark, Chris Olah, Benjamin Mann and Jared Kaplan were among numerous other Meta and Google DeepMind title-holders urging the US government to “support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.”
Iranian naval reconnaissance flagged by Anthropic
The report found that an Iranian-linked group, described as a “threat actor,” used the advanced AI Claude model to compile and analyse publicly accessible data to develop targeting recommendations against US naval forces in the region. This information included military photographs and commercial satellite photos, as the operator directed the chatbot to compile vulnerabilities research on shipboard’s software systems.
Elsewhere, Anthropic reported identifying an Iranian-linked operator for using AI tools to create an automated open-source intelligence identity-profiling system targeting Israeli governmental and non-government individuals and Jewish diaspora organisations.
A third instance revolved around 16 Claude accounts operated by two linked units associated with unidentified Iranian paramilitary and domestic security agencies being identified and banned.
One of them was described as a seven-department organisation with offices across Iranian provinces. Anthropic stated that it claimed to have an identity-record database of Iranian nationals. Another relied on Claude to work on a malicious browser extension used to mass-harvest user identities from top social network platforms.
Anthropic report warns of several national-security dangers
The AI company is now eyeing its mega IPO, having rejected the Pentagon’s demands over how its technology is used just months ago.
Earlier this year, Anthropic CEO Dario Amodei’s refusal to drop AI safeguards landed his company in the crosshairs with the Donald Trump administration. However, the company stood its ground by ultimately declaring not to work with the US Department of Defence rather than let its tech be used in a way that may “undermine, rather than defend, democratic values.” The company particularly found fault with the potential use of its AI tools, including Claude, for “mass domestic surveillance” and “fully autonomous weapons.”
The AI firm’s concerns were similarly laid bare in its more than 100-page-long report, which listed several allegations of global cyber operations and intelligence collection. The purported Iran case was just the tip of the iceberg.
Another such case highlighted in the report suggested that Chinese-speaking operators, expected to be based in the capital of central China’s Hunan province, used their AI model in cyber operations against about 50 organisations. These are even believed to have targeted foreign government networks.
Although some details still remain blurry, Anthropic pointed the finger at two university students in Hunan for supposedly using AI-driven workflows to search for previously undiscovered vulnerabilities and go as far as conducting parts of intrusions.
Yet another hacking group allegedly participated in phishing, hotel Wi-FI hijacking, and taking over WhatsApp operations of Ukrainian government officials. Anthropic said the group’s methods aligned with those of Russia-based actor Midnight Blizzard, which the US government once already linked to the country’s SVR foreign intelligence service.
Some other alleged surveillance threats recorded in the report indicated that China-linked operators could have used Claude to compile intelligence based on religion. Targets included Tibetan Buddhists, Taiwanese Christian leaders, Catholic cardinals, dissidents and activists.
Already under scrutiny for controlling organised religion and political dissent, China has been accused of conducting surveillance and press campaigns against critics abroad. The Asian nation has, however, dismissed such accusatory claims.
“As AI models become more widely used, providers will continue to acquire threat-relevant visibility into real-world use that even governments and intergovernmental organizations lack,” Anthropic warned in the lengthy September report. “We hope that sharing these early insights with the public helps inform governments, the industry, and the general public on the nature of these risks, and the safeguards that are necessary for ensuring the safe deployment of AI models.”