There’s been a lot of confusion following Anthropic’s launch of Claude Code Security. Let’s separate the reality from the noise and look at what is actually being disrupted.
What did Anthropic actually release? Claude Code Security is a capable AI tool designed to read a developer's codebase, find vulnerabilities, and suggest fixes. It is incredibly impressive.
The "Code to Cloud or Application Security" Illusion: Over the last few years, we watched a massive industry trend where major cybersecurity vendors were rushing to acquire tools to do exactly this. This category is known as Application Security. It is roughly a $20B market, which represents only about 5% of the total cybersecurity landscape.
Major players spent billions acquiring their way into this space — CrowdStrike buying Bionic; Palo Alto Networks acquiring Bridgecrew and Cider, Dig and more; Wiz pushing to Code; alongside pure-play names like Checkmarx and Snyk. The reality is that many of these legacy, rule-based code-scanning tools are about to be commoditized. AI models like Claude are simply better, faster, and cheaper at reading and reasoning through static code.
At SentinelOne, we intentionally chose not to acquire our way into this specific market. I have been vocal for the past few years that this exact segment was incredibly risky and ripe for AI disruption given AI’s code generation trajectory. More generally — the hope is code will be generated more securely from the get-go, with coding tools like OpenAI Codex, and Claude Code.
So, what does this mean for the future of cybersecurity? AI will subsume certain cybersecurity features of disjointed platforms (built by acquiring legacy narrow feature acquisitions) — specifically those based on manual review, audits, and pattern matching. I suspect that cloud misconfiguration posture management (aka CSPM) could be ripe for disruption - even though it is more nuanced and less structured than code security, so might not be as straightforward. However, active, real-time, and autonomous defense against sophisticated threat actors remains a highly defensible core of cybersecurity.
There is a massive difference between finding a bug in a static environment and stopping a live, adversarial cyberattack in real-time. Singularity was built on a fundamentally different premise: autonomous, AI-native security. We don't rely on manual audits or legacy pattern matching.
Active, real-time defense against threat actors is only getting more critical. The vendors built on autonomous, AI-native architectures will leverage this moment to get even stronger. The platforms relying on legacy patchworks have a difficult road ahead.
Speaking of that.. another super concerning LLM based cyberattack was uncovered: https://lnkd.in/gVYe_UGu

Jensen Huang and I share the same conviction: AI changes what the computing stack must provide for its behavior to be trustworthy.
In practice, that means a direct line between the agent runtime and the protected infrastructure beneath it, with visibility and enforceable control under every AI action. BlueField-4 and secure enclave architectures make it possible to secure the runtime from the hardware up, shrinking the surface that has to be defended down to the hardware boundary. Any AI, on any computer that supports it, gets safer by design.
Today, you can't responsibly hand consequential work to an AI agent without knowing three things: what it's doing, whose authority it's acting on, and whether the limits on that authority actually hold. Those guarantees can't rest on assuming a model following instructions. We've watched that fail too many times. The safeguards have to be built into infrastructure itself.
This is an important step, and there's more to build. Claims of trust have to be grounded in demonstrable behavior, and getting there will take sustained research and engineering across the stack. That is the work we at SentinelOne are committed to deliver.
Congratulations to the NVIDIA team on the launch of the NVIDIA Open Agent Safety Program. This is how increasingly capable AI becomes governable, and how organizations get comfortable entrusting it with work that matters.
Read more on my new blog 'The Endpoint': https://lnkd.in/g2PmwZqw
The name "fast16" appeared in the 2017 ShadowBrokers leak with one instruction to operators: "Nothing to see here — carry on." For twelve years, nobody knew what it was. Until now.
"...the most subtle form of sabotage ever seen in an in-the-wild malware tool: By automatically spreading across networks and then silently manipulating computation processes in certain software applications that perform high-precision mathematical calculations and simulate physical phenomena, Fast16 can alter the results of those programs to cause failures that range from faulty research results to catastrophic damage to real-world equipment..."
Tremendous work by Vitaly Kamluk and Juan Andres Guerrero-Saade (JAGS) to uncover fast16 — a cyber sabotage framework whose core components date to 2005. Five years before Stuxnet. The first of its kind.
Read on WIRED: https://lnkd.in/geVNkZk9
Read the blog: https://lnkd.in/gaFGG_WB
SentinelOne
