finance
Anthropic Restricts Access to Mythos AI Model Due to Cyberattack Risks
Anthropic has limited access to its new AI model, Mythos, following its discovery of over 10,000 vulnerabilities in open-source software. The model identified bugs, including a 27-year-old flaw in OpenBSD, and weaknesses in cryptography protocols like TLS and AES-GCM. Anthropic expects AI to enhance long-term cybersecurity defenses.
Apr 8, 11:20 AM·1m read1 source
app.buzzsumo.comAudio version
Tap play to generate a narrated version.
Anthropic, an artificial intelligence company, has restricted access to its new AI model named Mythos over concerns related to potential cyberattacks. The model was designed to detect software vulnerabilities and has identified more than 10,000 bugs in open-source projects. These findings prompted Anthropic to limit the model's availability to prevent misuse by malicious actors.
Mythos operates as a preview version and focuses on scanning code for security flaws. According to Anthropic, the model analyzed 15 popular open-source projects and uncovered the vulnerabilities within hours. The bugs include zero-day exploits, which are software flaws exploitable before developers can patch them.
the findings, Mythos identified several long-standing bugs, with many dating back 10 to 20 years.
The oldest detected was a 27-year-old vulnerability in OpenBSD, an operating system recognized for its security features; this bug has since been patched. The model also flagged weaknesses in widely used cryptography libraries, algorithms, and protocols, such as TLS, AES-GCM, and SSH.
Anthropic reported that detecting and patching such vulnerabilities traditionally requires specialized human expertise, which is both rare and costly.
AI models like Mythos could accelerate this process by scaling detection efforts. However, the company emphasized the need for caution in deploying such tools publicly.
stated that the discovery represents an early step in using AI for cybersecurity improvements.
The company anticipates that defending global cyber infrastructure will require years of effort. In the long term, AI is expected to contribute to more secure software through automated code hardening. " — Anthropic (via ZeroHedge) Access to Mythos remains limited as Anthropic evaluates risks and refines the model.
