Anthropic’s newest AI know-how, known as Mythos, is so highly effective at revealing software program vulnerabilities that the corporate is afraid to launch the mannequin publicly lest it fall into the palms of unhealthy actors.
The corporate, the developer behind the Claude AI chatbot, stated in a submit on its web site this week that the brand new instrument has already uncovered hundreds of weak factors in “each main working system and internet browser.”
Though that functionality might show to be a boon for safeguarding crucial techniques, it is usually stirring considerations that hackers might exploit Mythos to assault the IT infrastructure at banks, hospitals, authorities techniques and lots of different organizations.
Making ready for the “storm”
Slightly than releasing Mythos to the general public, Anthropic is sharing the tech with a choose group of main firms, together with Amazon, Apple, Cisco, JPMorgan Chase and Nvidia, to allow them to take a look at the mannequin and strengthen their very own techniques towards cyberattacks. Referred to as Undertaking Glasswing, the hassle is aimed toward serving to key firms harden their defenses earlier than hackers get entry to Mythos or comparable AI fashions, in response to Anthropic.
On the identical time, safety consultants stated, the considerations round Mythos attest to the risks of AI whether it is weaponized for hurt.
“What we have to do is have a look at this as a wake-up name to say, the storm is not coming — the storm is right here,” Alissa Valentina Knight, CEO of cybersecurity AI firm Assail, advised CBS Information. “We have to put together ourselves, as a result of we could not sustain with the unhealthy guys when it was people hacking into our networks. We actually cannot sustain now in the event that they’re utilizing AI as a result of it is a lot devastatingly quicker and extra succesful.”
Mythos’ capabilities are additionally sparking concern amongst federal officers. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell met with high financial institution CEOs in a closed-door assembly on Tuesday to debate Mythos and different rising cybersecurity dangers stemming from AI. Anthropic additionally briefed senior U.S. authorities officers and key business stakeholders on Mythos’s capabilities, CBS Information has discovered.
Individually, IMF Managing Director Kristalina Georgieva stated in an interview set to air Sunday on “Face the Nation with Margaret Brennan” that the world doesn’t have the flexibility “to guard the worldwide financial system towards huge cyber dangers.”
“The dangers have been rising exponentially,” Georgieva stated. “Sure, we’re involved. We’re very eager to see extra consideration to the guardrails which are essential to guard monetary stability on the earth of AI.”
Anthropic did not return a request for remark. In its submit, nevertheless, the corporate underscored the dangers of misusing instruments like Mythos. “The fallout — for economies, public security, and nationwide safety — may very well be extreme,” the corporate stated.
The weakest hyperlink
Such stark warnings masks one other troubling actuality: Hackers have already got entry to superior AI fashions and are utilizing them for a spread of malign functions, together with to create autonomous “brokers” able to finishing up assaults with out human intervention.
Such assaults vary from spreading malware and executing identification theft scams to producing deepfake movies and launching ransomware assaults, in response to cybersecurity consultants.
“AI-enabled tooling has empowered even low-skilled menace actors to execute high-speed, high-volume operations, while superior adversaries are utilizing AI to sharpen precision, scale automation and compress assault timelines,” PwC stated in a latest report.
“The time between the general public launch of a brand new functionality by an AI firm and its weaponization by menace actors shrank dramatically [in 2025], a development we assess will seemingly speed up in 2026,” the administration consulting agency added,
Different AI instruments, whereas not but as efficient as Mythos in exposing the mushy underbelly in software program, are already amplifying the dangers to shoppers, companies and governments. As an example, hackers are tapping AI to sharpen so-called phishing assaults aimed toward prying unfastened confidential info, stated Zach Lewis, the chief info officer on the College of Well being Sciences and Pharmacy in St. Louis.
“It has been used to essentially script these dialogues, these conversations, these phishing emails, to particular individuals — and actually customise them to make them much more troublesome to detect and determine if these are pretend or not,” he advised CBS Information.
“As soon as [Mythos] drops, we’ll see much more vulnerabilities, most likely much more assaults,” Lewis stated. “Cyberattacks are positively going to extend till we get to some extent the place we’re patching up all these vulnerabilities nearly in actual time.”
AI is more practical than people at discovering software program bugs as a result of it could actually shortly scan hundreds of traces of code and detect issues, one thing persons are not essentially good at, Knight defined.
“People are the weakest hyperlink in safety,” Knight famous. “People have the flexibility to make errors after we’re writing code. It is attainable for vulnerabilities in supply code to have by no means been discovered by people.”
On model for Anthropic?
Some safety consultants questioned the motives behind Anthropic’s incremental method to rolling out Mythos, speculating that the restricted launch may very well be aimed toward stirring intrest from different potential clients.
In the meantime, each Anthropic and rival OpenAI are anticipated to launch preliminary public choices by the top of the yr, in accordance to the Wall Avenue Journal — a attainable incentive to drum up headlines, stated Peter Garraghan, founder and Chief Science Officer at Mindgard, an AI safety platform.
“I think Anthropic could also be utilizing this as a advertising and marketing ploy, maybe in the direction of IPO,” he stated.
Anthropic has sought to tell apart its model from OpenAI and different rivals by publicly emphasizing AI security, highlighting its guardrails for retaining the know-how in line. Anthropic’s resolution to carry off on releasing Mythos and launching Undertaking Glasswing aligns with that picture, famous Columbia Enterprise Faculty advertising and marketing lecturer Malek Ben Sliman.
“When going through the robust selections, Anthropic has really been true to its values,” he stated. Curating the discharge of Mythos “does permit them to look to be the protectors of this accountable AI, but it surely additionally is a superb advertising and marketing and promoting instrument.”
