Security vendor Echo has launched Echo Frontier, a service that turns zero-day vulnerabilities found by AI models into tested patches, delivered to customers before the flaws are publicly disclosed.
Echo's case rests on how quickly offensive capability is spreading. It cites Anthropic research published on 29 September, in which Zhipu AI's open-weight GLM-5.3 produced working end-to-end exploits in 50 of 410 attempts on the ExploitBench benchmark, against 56 for Claude Mythos Preview. Unlike Mythos Preview, GLM-5.3 can be used by anyone, so unknown flaws can be found and exploited before a CVE, scanner signature or upstream patch exists.
Fixing is the harder half. Echo points to research from 1Password's Off-by-1 Labs which found that 54% of 6,080 AI-generated patches either failed to fix the vulnerability, introduced a new one, or both.
"The old assumption was that a vulnerability becomes public, gets a CVE and a fix, and then enterprises respond, but that sequence no longer holds," said Eilon Elhadad, co-founder and CEO of Echo.
Frontier builds on Echo's existing work vetting, patching, rebuilding and delivering open-source software. Its process runs from discovery, using model-driven analysis and human-led research, through validation against a working proof of concept, to notifying maintainers as an authorised CVE Numbering Authority. Echo then develops and backports a fix, tests it through the full application in a cloud environment, and delivers it privately through the customer's existing JFrog or Nexus setup. Customers move to the official upstream fix once it lands, and receive an evidence trail for compliance.
In one recent case Echo found a high-severity denial-of-service flaw in OpenEXR, validated it and protected customers in about two days, roughly two months before an upstream fix was available.
Echo Frontier is generally available now.
To stay across the latest in cloud, AI and enterprise tech analysis from Compare the Cloud, subscribe to our weekly newsletter at https://www.comparethecloud.net/newsletter
