Security practitioners are pushing back against alarms that Anthropic's unreleased Mythos AI model would unleash a wave of hacking, calling the response overblown one month after launch.
Practitioners Dial Back Mythos Panic
The hacking risks tied to Mythos look smaller than governments first feared, Reuters reported Wednesday. At launch in Apr., Anthropic said the model had uncovered thousands of software flaws spanning every major operating system and browser.
Officials in several countries met with banks to gauge exposure, and the White House by early May was weighing rules on how labs release new models after safety testing.
Inside the cybersecurity field, the reaction has been calmer. "I think there's a really big communication gap between practitioners and policymakers," Isaac Evans, founder and CEO of software security firm Semgrep, told Reuters. The model is "a real technical advance," he added, though the public response "is not substantiated by what we actually know."
Also Read: Claude Mythos AI Built Working Exploits Across 50 Cloudflare Repos, Then Refused To Demo
Experts See Measured Risk
The bigger problem is not finding bugs but triaging them. One vulnerability researcher with early access said AI has surfaced more flaws than teams can handle for months, with validation and patching the real bottleneck.
Mythos lowers the barrier to entry because it produces results from weaker prompts than earlier models required.
Anthony Grieco, senior vice president and chief security and trust officer at Cisco, pointed to faster code scanning and fewer false positives, helping defenders focus on the most pressing risks. Mythos also carries fewer guardrails than earlier releases.
Cynthia Kaiser, a former senior FBI cybersecurity official now at security firm Halcyon, said most attacks still do not depend on AI. "Our adversaries have gotten really good without AI," she said, noting that ransomware crews are now hitting victims in under an hour.
Project Glasswing Background
Anthropic launched Project Glasswing on Apr. 7, giving select organizations access to the Claude Mythos Preview for defensive cybersecurity work, with partners including Apple, Microsoft, Google, AWS and CrowdStrike. The Pentagon labeled Anthropic a supply-chain risk in March, even as the NSA reportedly kept using Mythos Preview. The White House in late Apr. rejected a plan to widen the partner list from roughly 50 firms to about 120.
Read Next: BitMine Buys 71,672 ETH As Tom Lee Calls $2,200 Dip A Bargain
