Houthis used Anthropic AI to try to build ballistic missiles

Yemeni militants sought to use Anthropic’s Claude AI model to develop guidance, control and navigation software for its ballistic missiles, the US company said.

In its intelligence threat report released on Thursday, Anthropic did not name the Iranian-backed Houthis, but it said it “identified a cell of threat actors based in northern Yemen” — much of which is controlled by the Houthi rebels — “running three weapons development programmes”.

The study highlights the growing risks of advanced AI falling into the wrong hands, as the technology continues to improve rapidly.

The same Anthropic report had also suggested that scientists made multiple attempts to use Claude to research biological weapons.

The US AI labs that are currently at the frontier of development face a constant struggle to ensure that unauthorised actors cannot use it to attack America and its allies.

Anthropic said the Yemeni cell sought to develop weapons including a multi-stage ballistic missile, a multi-variant missile, and a guided rocket that used a “commodity phone-class flight computer”.

“The actors used Claude Code in place of human software engineers to develop the guidance, navigation, and control software that steers and stabilises a flying vehicle,” Anthropic said in its report.

Thick black smoke and flames rise from vehicles and structures in a desert area after a missile strike, with scattered debris visible.
Thick black smoke and flames rise from vehicles and structures in a desert area of Yemen after a missile strike this week © Houthi Media Center/Getty Images

“They used Claude to integrate an open-source autopilot on to a phone-class flight computer, writing the control and position estimation software, tuning the control settings, running a firmware build pipeline, and performing a flight simulation.”

Anthropic said that its safeguards blocked “many of their requests, but not all of them,” adding that the “actors used a variety of tactics to evade” the measures.

They did this by “hiding their goals and the products the software was meant for, and they split their work across multiple sessions so no single session revealed their full intent”.

The Houthis, who are one of the most potent members of Iran’s so-called axis of resistance and are designated a terrorist group by the US, this week launched a blitz along Yemen’s Red Sea coast, capturing a strategic port and advancing on the Bab al-Mandeb Strait, a vital waterway for international shipping.

The rebel offensive comes as they have stepped up attacks on Saudi Arabia’s energy infrastructure and reignited Yemen’s more than decade-long civil war. The Houthi advance on the Bab al-Mandeb, which connects the Red Sea to the Gulf of Aden and the Indian Ocean, helped push up the price of Brent crude to more than $105 a barrel.

The battle-hardened Houthis, which have controlled the Yemeni capital Sana’a and much of Yemen’s populous north since 2014, have been developing their missile capability, analysts say, and receive weapons and technology from Iran.

Their advance towards the Bab al-Mandeb has dealt a significant blow to Saudi Arabia, which intervened in Yemen’s civil war to lead an Arab coalition against the Houthis in 2015, and boosted Iran as it seeks to push up energy prices as part of its battle with the US.

The Houthis had largely stayed out of the Iran war, with the exception of firing several salvos of missiles and drones at Israel.

A 2022 truce with Saudi Arabia and the Yemeni government, which Riyadh backs, had held until July when the Houthis declared that they were imposing a blockade on Saudi ports and began attacking Saudi tankers.

The Bab al-Mandeb has become a vital route for Saudi oil exports since Iran throttled shipping through the Strait of Hormuz after the US and Israel launched their war against Iran in February.

Anthropic said the “actors” in northern Yemen “carried out a sustained effort to develop guided weapons, including using Claude to design guidance software”.

“We do not have evidence the actors succeeded in fielding an operational device; but they did test-fire a guided rocket,” it said in its report. “This field test appears to have failed: within hours, the actors returned to Claude to work out why it failed.”

Anthropic declined to comment on whether the “threat actors based in northern Yemen” were indeed the Houthis, but few other groups fit that description.

The AI company said it banned accounts associated with the actors and shared threat information with public- and private-sector partners to mitigate risks posed by the actors.

“Nevertheless, we have evidence that the actors had already built an offline simulation toolkit that does not rely on Claude or other engineering computing environments,” it said.

Additional reporting by Madhumita Murgia

添加评论
点赞收藏
点踩分享查看原文
评论
?
参与讨论