Anthropic just dropped Project Glasswing — a big collaborative cybersecurity initiative with a shiny new model called Claude Mythos Preview that can find zero-day vulnerabilities at scale. Twelve major tech companies involved. $100M in credits. Found a 27-year-old flaw in OpenBSD. Impressive stuff.

But let's be real about what's happening here. Anthropic trained a model so capable at breaking into systems that they decided it was too dangerous to release publicly. So instead of sitting on it, they wrapped the reveal in a security initiative and turned a liability into a PR win. Smart move — and I'm not saying the security work isn't valuable, it clearly is. But the framing is doing a lot of heavy lifting.

The part that actually matters, though, is who benefits. Glasswing is for the big players. The companies with security teams, budgets, and the kind of infrastructure that gets invited to sit at the table with AWS, Microsoft, and Palo Alto Networks. What about the rest of us? The startups, the small SaaS shops, the indie developers running production systems on a shoestring?

The internet is a dark forest. That's not a metaphor anymore — it's becoming the literal reality. Bots, scrapers, automated exploit chains, credential stuffing, AI-generated phishing. A server goes up and within hours it's being scanned, fingerprinted, and probed by systems that don't sleep. Visibility equals vulnerability. And AI is making the attackers faster, cheaper, and more autonomous every month.

The ISC2 put it plainly — both offence and defence now operate at speeds beyond human intervention. The threats aren't people sitting at keyboards anymore. They're autonomous systems running campaigns end-to-end.

So what do we do about it?

Offensive security — but not the kind you're thinking

When I say offensive security, I don't mean red-teaming or penetration testing. I mean giving your systems the ability to fight back.

Picture an LLM that sits across your centralised logs — network traffic, database queries, user interactions, access patterns — and builds an understanding of what normal looks like for your system over weeks and months. Not just pattern matching against known signatures. Actually understanding the shape of healthy behaviour.

When something breaks the pattern, it doesn't just alert. It acts.

Disable a compromised account. Kill a service that's behaving strangely. Block a database connection that shouldn't exist. Create an incident with full context for a human to review. The response is proportional and immediate — not waiting for someone to check their phone at 3am.

The architecture is pretty straightforward:

graph TD
    A[Application Logs] --> D[Secure Isolated Log Store]
    B[Network Traffic] --> D
    C[Database Queries] --> D
    D --> F[Baseline Health Model]
    E[User Activity] --> D
    F -->|Anomaly Detected| G[LLM Analysis]
    G -->|Analyse & Plan| H{Threat Assessment}
    H -->|Low| I[Alert & Log]
    H -->|Medium| J[Restrict & Escalate]
    H -->|High| K[Disable & Isolate]
    I --> L[Human Review]
    J --> L
    K --> L

The key is that the logging and analysis layer has to be isolated and secured separately from the systems it's watching. If an attacker can compromise the thing that's watching them, the whole model falls apart. You centralise the logs, lock them down, and give the analysis layer specific, scoped abilities to pull levers — disable access, restrict traffic, shut down services. Nothing more.

Where biology gets interesting

I've been doing research with my project C302 — using a simulation of the C. elegans roundworm's neural network as a behavioural controller for LLM agents. The worm has 302 neurons. That's it. And with those 302 neurons it navigates its environment, finds food, avoids threats, and adapts its behaviour based on what's working.

In that research, we mapped simple feedback signals to biological synapses and let the neural simulation drive agent behaviour. The live connectome — receiving real-time feedback from the agent's environment — showed a clear improvement over one following a fixed trajectory (0.960 vs 0.867 test pass rate), even when the topology, signals, and rules were identical. The only variable was whether the system adapted to what was actually happening. Early days with a small sample size, but the direction is promising.

Now apply that thinking to security monitoring.

Imagine mapping a sudden spike in unusual user activity to the equivalent of a "salt" sensory neuron in the worm's circuit. That fires, and the downstream effect is the security system becomes more aggressive in its investigation — widening its search, correlating more signals, lowering its threshold for action. A pattern of repeated failed authentications from new IPs could map to a "touch" response — the system recoils, tightening access controls automatically.

This isn't rule-based. It's adaptive. The system develops a behavioural pattern that's learned from running in your specific environment, responding to your specific traffic patterns. That's a fundamentally different thing from a static set of if-then rules.

This has to be open

Glasswing is cool. Open-source frameworks like CAI are a good start. But we need much more.

The cost of AI-powered security monitoring needs to come down to the point where a three-person startup can run the same kind of defensive intelligence that a bank does. Not the same scale, obviously — but the same principles, the same tooling, the same ability to watch and respond.

That means open-source models trained for security analysis. Open standards for log formats and threat signals. Publicly available tooling for building the kind of centralised-logging-to-LLM-analysis pipeline I've described above. This can't be proprietary. It can't be locked behind enterprise contracts.

The dark forest doesn't care how big your company is. The bots scanning your infrastructure don't check your headcount before they attack. If we're serious about security in the age of AI, the defensive tooling has to be as accessible as the threats are pervasive.

I'd love to take the C302 work further in this direction — building an adaptive, biologically-inspired security agent that anyone can deploy. It's early days but the pieces are all there. The worm, the logs, the levers. Someone just needs to wire them together.