Starting out? Start with Cyber 101!
The past few weeks have been unusually busy for vulnerability disclosures and public security research. Some of the more widely discussed include PHP2Shell, Node2Shell and SignDrop, spanning web languages, shared software libraries and Windows. In several cases, exploitation attempts followed within days of disclosure. Taken individually, each is the kind of event security teams handle routinely. What interests us is the pattern. It may reflect a landscape that is becoming harder to defend, or it may be the early stage of a long overdue clean-up, in which flaws that have sat unnoticed for years are finally being found and fixed.
References
We lean towards the second reading, because vulnerability research is becoming faster and more affordable. OpenAI recently reported that a pipeline of autonomous AI agents, in a single run against the Linux kernel, produced working exploits for 948 distinct privilege escalation flaws along with a candidate patch for each. By its account, more than 100 of those had been independently patched or had fixes in staging from the wider kernel community within ten days. These are the company’s own figures and they await independent confirmation. If they hold, a clean-up on that scale is good news over the long term.
The same capability is available to adversaries, which is where the pace of patching starts to matter. An earlier OpenAI run with an older model produced 166 such exploits, and the company expects less capable models to manage the simpler task of developing an exploit once a flaw has been disclosed. Models of many kinds are already widely accessible, and OpenRouter’s usage rankings show seven of the ten most used last week came from developers other than OpenAI and Anthropic. We think it sensible to plan on the basis that a published flaw can be turned into a working exploit quickly, and by a wider range of actors than before.
References
For most organisations, the sensible response is a familiar one. A busier disclosure calendar rewards those with less to maintain, so it is worth reviewing which services need to face the internet. Administrative interfaces and remote management tools are well suited to sitting behind a VPN or identity-aware gateway. Whatever stays exposed is best kept on supported software, patched to a defined timescale and hardened through least privilege, segmentation and secure configuration baselines. A smaller footprint also lightens the load on the security, operations and development teams who absorb each new advisory.
Infrastructure and services that are working well are worth keeping. The aim is to adjust how they are exposed and maintained, using well-established fundamentals that are within reach of organisations of every size. The current pace of disclosure makes a good case for revisiting them calmly and on your own schedule.
If you would like to talk through what your organisation presents to the internet, we are always happy to have that conversation.
