entityOS Security
Project Glasswing confirmed what many have suspected: AI has crossed a threshold where it can find and exploit vulnerabilities in every major operating system and browser — autonomously, at scale. The open, anonymous internet is no longer a viable foundation for critical software. The answer is cryptographic identity for every connection, every actor, every request.
The Inflection Point
In April 2026, Anthropic's Project Glasswing announced that Claude Mythos Preview — a frontier AI model — had autonomously found thousands of critical zero-day vulnerabilities across every major operating system and web browser. Some had survived decades of human review and millions of automated security tests. This is the threshold moment: the cost of finding and exploiting software vulnerabilities has collapsed.
Frontier AI has reached a level where it surpasses all but the most skilled humans at finding and exploiting software vulnerabilities — and the capability is proliferating rapidly.
Every major OS, every major browser, every significant piece of infrastructure contains flaws. The question is no longer whether vulnerabilities exist — it is who finds them first.
Mythos Preview found a 27-year-old flaw in OpenBSD and a 16-year-old bug in FFmpeg that had survived five million automated test runs. Human review timescales are no longer adequate.
The model identified nearly all vulnerabilities and developed many exploits entirely without human steering. No operator fatigue. No oversight gaps. Constant, scalable attack surface scanning.
"AI models have reached a level of coding capability where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities... The fallout — for economies, public safety, and national security — could be severe."
Anthropic — Project Glasswing, April 2026The Core Problem
The internet's foundational assumption — that anonymous connections should be permitted until proven harmful — made sense when attacking systems required rare human expertise. It does not make sense when AI can scan, find, and exploit vulnerabilities at machine speed and scale.
Project Glasswing
Three examples from Anthropic's Frontier Red Team that illustrate why the old security model cannot hold. Each represents a class of vulnerability that AI will now find routinely — in your systems, before your team does.
A vulnerability that allowed an attacker to remotely crash any machine running OpenBSD — used to run firewalls and critical infrastructure — simply by connecting to it. Survived 27 years of expert human review.
A flaw in a single line of code in FFmpeg — used by innumerable pieces of software to handle video — that automated testing tools had hit five million times without ever detecting it. Mythos Preview found it immediately.
The model autonomously found and chained together multiple vulnerabilities in the Linux kernel — escalating from ordinary user access to complete control of the machine. No human steering required.
The Defender's Response
The same AI capabilities that make attacks more potent make defences more powerful. Project Glasswing is proof: AI used proactively found and patched thousands of flaws before attackers could exploit them. The answer is not to slow AI — it is to point it in the right direction, behind cryptographic identity walls.
The Glasswing coalition — AWS, Apple, Cisco, Google, Microsoft, NVIDIA and others — is using Mythos Preview to find and patch vulnerabilities in their own codebases before adversaries can exploit them. This is the correct use of AI-augmented cyber capability: defender advantage through speed.
AI-driven attack tools can only exploit what they can reach. Cryptographic identity — enforced via IP allowlisting and mTLS — closes the attack surface before the scan can begin. A service that requires a certificate cannot be reached by an anonymous AI probe.
When AI can find and exploit a vulnerability in hours, long-lived credentials become liabilities. Short-lived credentials — rotated automatically every 1–24 hours — limit the window of exposure for any credential compromise to hours, not months.
The entityOS Architecture
entityOS Security implements the closed internet in three progressive stages — each adding a deeper layer of cryptographic identity that AI-driven attacks cannot bypass.
What This Means For You
The Glasswing threshold does not only affect large technology companies. Any organisation running software — which is every organisation — is now operating in an environment where AI-augmented attackers can find and exploit vulnerabilities at a scale and speed that human security teams cannot match.
Every internal API, every cloud service, every third-party integration is a potential entry point. AI tools do not discriminate by organisation size — they scan everything reachable. The immediate priority is to make your services unreachable to anonymous callers.
Blockchain-based identity provides strong proof of ownership and governance rights. But the network connections between wallets, nodes, and APIs are still exposed to the open internet threat model. On-chain identity must extend to mTLS-protected service connections and KERI-anchored communication channels.
AI-assisted development generates code faster than human review can evaluate it. AI-assisted security scanning can now audit that code more thoroughly than any human team. The two must be paired: every codebase that uses AI to build must also use AI to secure — and the resulting software must run behind cryptographic identity walls.
The closed internet requires every actor to carry cryptographic identity — not just organisations and services. Individuals will need portable, self-sovereign identity credentials to participate in a system that requires proof of who they are at every connection. KERI-based self-certifying identifiers provide the foundation for individual digital identity that is independent of any central authority.
The Imperative
Project Glasswing is not a warning about the future. It is a description of the present. The transition to cryptographic identity for every connection is not a long-term aspiration — it is an immediate operational requirement.
The appropriate response to the Glasswing moment is to accelerate, not wait. Stage 1 — IP allowlisting — can be deployed today. Stage 2 — mTLS with X.509 identity — should be in planning this quarter. The organisations that move now will be the ones that maintain a defender advantage as AI capabilities continue to advance.