The vulnerabilities business is being reorganised around a shrinking clock. A suspected supply chain attack pushed malicious JavaScript onto sites operated by CNN, Fox News, the BBC, Salesforce and others; Google Project Zero decided to abandon its long-standing 90+30 disclosure model for shorter fixed deadlines, citing AI-accelerated discovery and exploitation; and a vendor launched tooling aimed squarely at managing how automated defenders act. Different stories, one thread: the time defenders can assume they have is getting shorter, and both the rules and the tools are being rewritten in response.
The window is the vulnerability now
The Optimizely incident, as reported by BleepingComputer, is the clearest illustration of how little slack exists between a weakness in a supply chain and its use against ordinary visitors. The affected sites include Zoom, CNN, Fox News, the BBC, Salesforce and GitLab, and the injected scripts were designed to steal cryptocurrency and payment information and to exploit vulnerabilities in iOS devices. No single organisation had to be breached in the conventional sense; the third-party service in the path did the distributing. For US technology companies, the lesson is structural rather than tactical: the period in which a compromised dependency can be detected and removed is measured against how quickly the payload starts working on visitors, not against how long an internal patch cycle takes.
Disclosure policy admits the race
Google Project Zero's decision, also reported by BleepingComputer, to replace its "90+30" policy with shorter fixed deadlines is an admission from the most prominent disclosure shop in the industry that the old timetable assumed a pace that no longer holds. The stated reason was AI-accelerated vulnerability discovery and exploitation. That matters for US companies because Project Zero's deadlines have functioned as an informal industry metronome: vendors plan remediation around them, and a shorter clock compresses the interval in which a fix can be developed, tested and shipped without a public countdown running out. The policy change does not create the pressure, but it formalises it. Organisations that treat disclosure deadlines as an external courtesy rather than an operational constraint will find the gap between the two closing.
Defence moves toward machine speed
The third story, from SiliconANGLE, describes Cogent Security's launch of DefenseOS, an operating system for AI agents conducting defensive cybersecurity work inside the enterprise, alongside Agent Policies and Agent Automations. Agent Policies is intended to give the company's AI agents a deep understanding of the enterprise environment, while Agent Automations covers the actions those agents take. This is the same clock problem approached from the vendor side: if discovery and exploitation are accelerating, the response cannot be a queue of tickets awaiting human triage. The emphasis on policies is telling. Automating defensive action inside a corporate network raises questions about what the agents are permitted to do and how that permission is bounded, and the product framing suggests the vendor sees governance as the precondition for speed rather than an afterthought.

