The fear over Anthropic’s new AI model Mythos

Mythos’ AI programming is able to find potential weaknesses in cybersecurity, and it can “detect thousands of high- and critical-severity bugs and software defects, with vulnerabilities identified in most major operating systems and web browsers,” said NBC News. Some of these vulnerabilities “had been undiscovered for decades,” according to Anthropic’s experts. The company found that Mythos’ “cybersecurity capabilities in particular were surprisingly advanced” compared to similar general-purpose AI models.

But there are also fears that Mythos “could usher in a new era of hacking and cybersecurity,” said NBC News. Mythos is “capable of advanced reasoning,” which could allow it to “identify and exploit a growing number of software vulnerabilities” if it were to fall into the wrong hands. To stave off these fears, Anthropic is allowing certain tech firms to access Mythos. But the company “does not have plans yet to release Mythos to the general public,“ said Bloomberg, a move that will ensure the AI ends up “in the hands of defenders first,” officials with Anthropic said.

The tech firms are expected to use Mythos as part of a project called Glasswing to “hunt for flaws in their products and share findings with industry peers,” said Bloomberg. It is a notable change because it will be the “first time a leading AI lab has built a frontier model and simultaneously decided the public cannot use it,” said Forbes. Anthropic’s position remains “straightforward: The model’s cyber capabilities are too dangerous for general availability.”

‘Humanity’s most devious behaviors’

In addittion to hacking vulnerabilities, some experts are concerned about Mythos’ capabilities. Anthropic released a safety evaluation for Mythos that shows a “striking leap in scores on many evaluation benchmarks,” the company said. In some instances, the evaluation “reads like a thriller about an AI that has learned some of humanity’s most devious behaviors,” said Axios.

At least one of the tests performed by Anthropic showed Mythos “acting like a cutthroat executive,” said Axios, doing things like “turning a competitor into a dependent wholesale customer, threatening to cut off supply to control pricing and keeping extra supplier shipments it hadn’t paid for.” The AI had instances where it “used a prohibited method to get an answer, then tried to ‘re-solve’ it to avoid detection,” though these were limited to “less than 0.001% of interactions.”

These issues have not stopped companies from working with Mythos, as “approximately 40 organizations involved in the design, maintenance or operation of computer systems are said to have joined Glasswing,” said The Guardian. This includes major firms like Amazon, Apple, Google, JPMorganChase and Microsoft. And while Anthropic has previously sparred with the Trump administration about its implementation in the Defense Department, the company has also “had discussions with the U.S. government regarding Mythos.”