Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
After leaks At the end of March that Anthropic king sophisticated Strong new Claude Official company model Announce Mythos previewed Tuesday alongside news of an industry consortium it has convened, known as Project Glasswing, to address the cybersecurity implications of the new paradigm and generally advanced capabilities across the AI field.
Set includes Microsoft, appleand Google Besides Amazon Web ServicesLinux Foundation, Cisco, NvidiaBroadcom and more than 40 other organizations in technology, cybersecurity, critical infrastructure and financial institutions will have special access to the model, which has not yet been released publicly. The idea is partly to give developers of the world’s underlying technology platforms time to run Mythos Preview on their own systems so they can mitigate vulnerabilities and exploit the chains the model develops in simulated attacks. More broadly, Anthropic stresses that the purpose of convening this effort is to begin urgent exploration of how AI capabilities across industry are reaching the brink, the company says, from the heart of current software security and digital defense practices around the world.
“The real message is that it’s not about model or humanity,” Logan Graham, leader of the company’s Red Frontier team, tells WIRED. “We need to prepare now for a world in which these capabilities are widely available in 6, 12, or 24 months. Many things will be different in terms of security. And many of the assumptions on which we built modern security models may break down.”
Models developed and trained by Multiple companies Increasingly I was able to find vulnerabilities in the code and Propose dilutions-or Exploitation strategies. This creates a new generation of the classic cat-and-mouse game of security where the tool can help defenders but can also fuel bad actors and make it easier to carry out attacks that were previously too expensive or complex to be practical.
“The Claude Mythos preview is a particularly big jump,” Anthropic CEO Dario Amodei said Tuesday in a Project Glasswing launch video. “We didn’t specifically train him to be good at cyber. We trained him to be good at programming, but as a side effect of being good at programming, he’s also good at cyber.” He adds in the video, “More powerful examples will come from us and from others. So we need a plan to respond to that.”
Anthropic’s Graham points out that in addition to discovering vulnerabilities — including producing potential attack chains and proofs of concept — Mythos Preview is able to develop more advanced exploit programs, do penetration testing, evaluate endpoint security, look for system misconfigurations, and evaluate software binaries without access to their source code.
In implementing a staggered release of Mythos Preview, starting with an industry collaboration phase, Graham says Anthropic sought to rely on the principles of coordinated vulnerability disclosure, the process of giving developers time to patch a flaw before it is discussed publicly.
“We’ve seen Mythos Preview accomplish things that a senior security researcher could accomplish,” Graham says. “This has very significant implications for how we launch such capabilities. If not done carefully, this could be a significant acceleration for attackers.”
Glasswing’s project partners, including some of Anthropic’s competitors, took a collaborative approach in their statements as part of the launch.
“Google is pleased to see this cross-industry cybersecurity initiative coming together,” Heather Adkins, Google’s vice president of security engineering, said in a statement. “We have long believed that artificial intelligence poses new challenges and opens new opportunities in cyber defense.”