Thirty two new positions, with salaries reaching hundreds of thousands of dollars, are being opened by Anthropic, which is seeking experts to identify how artificial intelligence systems could be exploited for the construction of explosives, the development of nuclear weapons, cyberattacks and other forms of mass destruction. However optimistic one may be about artificial intelligence, the nature of these vacancies is enough to make anyone stop and wonder.
Anthropic's job listings read more like a catalogue of global threats than typical openings at a technology company, according to Axios.com. In total, the company is recruiting for 32 roles, all sharing a common goal: the early detection of ways in which someone could exploit AI systems to cause real world destruction.
From explosives to nuclear weapons
Anthropic is hiring analysts specialising in chemicals and explosives, nuclear and radiological weapons, financial fraud, cybercrime and other forms of malicious use of artificial intelligence.
One of the listings states that the candidate will play a critical role in protecting against the misuse of AI systems for radiological and nuclear harm.
The new employees will be asked to think like those attempting to bypass safety filters, extract dangerous information from the models and use them to build weapons or carry out attacks.
Salaries approaching $300,000
The company is prepared to pay a premium for this expertise, with compensation for several of the positions ranging from the mid to high $200,000s annually.
The level of the salaries reflects the weight AI companies now attach to the risk of their most powerful models becoming tools in the hands of criminals, terrorist organisations or state actors.
An Anthropic spokesperson said that a core requirement of responsible development is ensuring that models do not provide potentially dangerous information. As the spokesperson explained, the company systematically hires experts in sensitive fields who understand both the nature of the threats and the ways in which artificial intelligence could amplify them.
Searching for weaknesses before models are released
The experts' work is not limited to programming. They are expected to test systems before release, attempt to bypass safety mechanisms and identify gaps that malicious users could exploit.
Anthropic says it already employs hundreds of staff working exclusively on safety issues. These teams carry out continuous stress testing of the models and correct the weaknesses they identify.
For this reason, the positions require genuine experience in fields such as biology, nuclear technology, explosives and cybersecurity. Knowing how an AI model works is not enough; an understanding of how it could be used in the real world to cause harm is also required.
Amodei's warnings of attacks with millions of casualties
The Chief Executive Officer of Anthropic, Dario Amodei, has repeatedly warned that artificial intelligence could dramatically enhance the capabilities of malicious actors.
In a January essay, he described biological attacks as one of the most worrying scenarios. He noted that he does not consider it certain such an attack would take place as soon as it becomes technically feasible. He estimates, however, that as the number of people with access to powerful models grows and as the years pass, so does the risk of a major attack with millions of casualties.
Anthropic has frequently faced criticism for adopting an overly catastrophist stance towards artificial intelligence. The new hires, however, show that the company is not limiting itself to warnings but is investing significant sums to address the risks it describes.
The clash with the Pentagon
At the start of the year, Anthropic clashed with the US Department of Defense over the potential use of its technology in mass surveillance systems and autonomous weapons.
The disagreement highlighted the question of who decides how the most powerful AI models may be used, particularly when the state, the military and private companies have different priorities.
OpenAI also seeking biological risk experts
Anthropic is not the only company strengthening its safety teams. OpenAI is also seeking a researcher specialising in biological and chemical risks, offering an annual base salary of $295,000 to $445,000.
As the models grow more powerful, the industry's largest companies are attempting to pre-empt risks that until recently seemed closer to science fiction scenarios.
The companies themselves are being asked to set the rules
Expertise is increasingly moving towards the private sector and away from governments and public authorities. Combined with the absence of a coherent international regulatory framework, the very companies building the most powerful models are being called on to decide which risks are acceptable, which filters are considered sufficient and which uses must be prohibited.
The contradiction is plain: AI companies are developing systems that promise to change the world while, at the same time, hiring teams of experts to ensure those same systems are not used to destroy it.
