On Tuesday, Anthropic unveiled its forthcoming artificial intelligence model, dubbed Claude Mythos, which has exhibited remarkable efficacy in uncovering software vulnerabilities.
This innovative model has revealed thousands of weaknesses in widely used applications, many of which lack existing patches or fixes.
In response, the San Francisco-based AI startup has initiated a collaboration with cybersecurity experts to enhance protection against potential hacking threats.
“We possess a novel model that we have specifically chosen not to release to the public,” stated Mike Krieger of Anthropic Labs during the HumanX AI conference held in San Francisco.
Rather than making Mythos broadly accessible, Anthropic is enabling cybersecurity specialists and members of the open-source community to leverage this technology as a form of defensive armament, a strategy Krieger articulated as “arming them ahead of time.”
Advancements in AI capabilities have sparked apprehension about potential misuse by hackers who could harness such tools to decipher passwords or breach encryption protocols designed to safeguard sensitive data.
The most antiquated vulnerabilities unveiled by Mythos date back a staggering 27 years, with none having been previously detected by their developers, as per Anthropic’s findings.
Mythos, the latest iteration from Anthropic’s Claude series, faced scrutiny following a recent code leak, prompting the startup to issue a blog post highlighting the unprecedented cybersecurity threats it may pose.
“AI models have achieved a level of coding proficiency that allows them to outpace all but the most adept humans in identifying and exploiting software vulnerabilities,” Anthropic declared in its blog.
“The ramifications could be dire—for economies, public safety, and national security.”
The vulnerabilities exposed by Mythos were, in many instances, intricate and exceedingly difficult to identify without the aid of AI technology, according to Anthropic.
As an illustration, the model identified a previously unnoticed flaw in a video application that had undergone more than 5 million tests by its developers.
Project Glasswing
As a precautionary measure, Anthropic has disseminated a version of Mythos to cybersecurity firms such as CrowdStrike and Palo Alto Networks, as well as tech giants Amazon, Apple, and Microsoft, in an initiative referred to as “Glasswing.”
Networking leaders Cisco and Broadcom have joined this endeavour, alongside the Linux Foundation, which champions the open-source Linux operating system.
“The magnitude of this work necessitates collaboration; it is too vital and pressing to undertake alone,” asserted Cisco’s chief security and trust officer, Anthony Grieco, in a joint announcement regarding Glasswing.
“AI capabilities have transcended a critical threshold, fundamentally altering the urgency to safeguard essential infrastructure from cyber threats, and there is no turning back.”
Approximately 40 organisations, involved in the design, maintenance, or operation of computer systems, are reported to have joined the Glasswing initiative.
Project contributors are expected to exchange their findings related to Mythos, with Anthropic committing around $100 million in computing resources to support the mission.
Preliminary applications of AI models have demonstrated their potential to identify and rectify software and hardware vulnerabilities at an unprecedented pace and scale, according to Grieco.
“The interval between the discovery of a vulnerability and its exploitation by adversaries has drastically diminished—what once required months can now transpire within minutes, thanks to AI,” remarked CrowdStrike’s chief technology officer, Elia Zaitsev.
“The Claude Mythos Preview showcases the capabilities now attainable for defenders at scale, while adversaries will inevitably seek to exploit similar advantages.”
Despite a White House decree issued in February to sever all contracts with the startup, Anthropic has engaged in discussions with the U.S. government regarding Mythos. That directive has since been paused by a federal court judge, as the startup navigates its legal challenge.
Source link: Fox21online.com.
