Thus far, the world has been focusing on AI agents that perform simple tasks, but it appears that some of the top labs are thinking of taking things even further.
Anthropic says that it expects fully-AI employees to begin working in real-world companies in a year. The assertion was made by Jason Clinton, Anthropic’s chief information security officer in an interview with Axios. Managing these AI identities will require companies to reassess their cybersecurity strategies or risk exposing their networks to major security breaches, Clinton said.
Unlike AI agents, which focus on a specific, programmable task, AI employees will get a lot more done. These AI employees would have their own “memories,” their own roles in the company and even their own corporate accounts and passwords. Such AI employees will have far more autonomy than AI agents have today.
But giving these AI employees autonomy like regular employees could also give rise to new challenges. Providing access to these employees to company systems through passwords and logins could introduce new security challenges at companies, Clinton said. Also, having these AI employees could also complicate how blame is assigned in case an employee messes up — Clinton gave an example of a rogue AI employee that messes up a company’s codebase over a course of two weeks. “In an old world, that’s a punishable offense,” he said. “But in this new world, who’s responsible for an agent that was running for a couple of weeks and got to that point?”
The fact that AI employees are being discussed in such detail shows that Anthropic believes that their deployment could well be on the horizon. Anthropic has had aggressive-sounding timelines to AI progress — last month, CEO Dario Amodei had said that 90 percent of all code could be written by AI in the next 3-6 months, with a 100 percent being possible in a year. Anthropic now seems to have a similarly aggressive timeline for AI employees. It does stand to reason that if AI agents continue to grow at the same pace as the rest of AI, they really could become enormously capable quite quickly, and become full-fledged employees as opposed to being mere agents. But if these AI employees do enter the workforce in a year, they could compound the job displacement that already seems to be poised to happen thanks to AI.