Anthropic finds more than 10,000 critical vulnerabilities in software worldwide

Anthropic has launched Glasswing, a closed cybersecurity project that hunts for vulnerabilities in critical software worldwide. In its first month it found more than 10,000 vulnerability candidates, of which 1,094 are confirmed as high-severity. The numbers show how serious the security problem is for the software that holds up the global internet.

AI-processed from TNW; edited by Hamidun News
Anthropic finds more than 10,000 critical vulnerabilities in software worldwide
Source: TNW. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTNW ↗
Editor-in-Chief
Key metric #1Critical bugs in system libraries and network services
Key metric #2Authentication and authorization problems
Key metric #3Memory leaks, buffer overflows, race conditions

Anthropic has revealed the scale of a problem the information security world habitually underestimates. Its cybersecurity project Glasswing found more than 10,000 critical vulnerabilities in systemically important software in a single month. Of these, 1,094 real high-severity threats are confirmed.

The scale of the problem

Glasswing was launched as a limited Anthropic initiative. In its first month, the system flagged more than 10,000 vulnerability candidates in critical software used around the world. 1,726 true positives have been confirmed. Another 1,094 are classified as high- or critical-severity, meaning they need urgent fixes.

This is not just a statistic. It is an admission that the software underpinning the internet and critical infrastructure is full of serious holes that nobody has systematically closed.

What vulnerabilities were found

  • Critical bugs in system libraries and network services
  • Authentication and authorization problems
  • Memory leaks, buffer overflows, race conditions
  • Misconfigurations in popular open-source projects
  • Input-handling vulnerabilities

Anthropic uses Claude for large-scale code analysis. The model can quickly scan large codebases and identify potential problems, work that would take a human audit years. AI does not replace the security expert, but it significantly speeds up discovery.

The problem of fix speed

The main catch: finding a vulnerability is only half the job. It still has to be fixed, tested, and the patch deployed. If AI finds vulnerabilities faster than developers can close them, security debt grows.

Anthropic clearly hopes that publicly disclosing these figures will push the global developer community to take code security more seriously. But there is a risk of the opposite effect: a figure of 10,000 vulnerabilities may frighten people instead of motivating them.

What it means

AI systems can now audit code at a scale and speed beyond human reach. That changes the security landscape. But the developer community must speed up its fixes for the identified problems, or we will live with a growing number of known but unpatched vulnerabilities in critical software.

Story

Key Questions Answered

What are the key details on The scale of the problem?

Glasswing was launched as a limited Anthropic initiative. In its first month, the system flagged more than 10,000 vulnerability candidates in critical software used around the world. 1,726 true positives have been confirmed. Another 1,094 are classified as high- or critical-severity, meaning they need urgent fixes. This is not just a statistic. It is an admission that the software underpinning the internet and critical infrastructure is full of serious holes that nobody has systematically closed.

What vulnerabilities were found?

Critical bugs in system libraries and network services Authentication and authorization problems Memory leaks, buffer overflows, race conditions Misconfigurations in popular open-source projects Input-handling vulnerabilities Anthropic uses Claude for large-scale code analysis. The model can quickly scan large codebases and identify potential problems, work that would take a human audit years. AI does not replace the security expert, but it significantly speeds up discovery.

What are the key details on The problem of fix speed?

The main catch: finding a vulnerability is only half the job. It still has to be fixed, tested, and the patch deployed. If AI finds vulnerabilities faster than developers can close them, security debt grows. Anthropic clearly hopes that publicly disclosing these figures will push the global developer community to take code security more seriously. But there is a risk of the opposite effect: a figure of 10,000 vulnerabilities may frighten people instead of motivating them.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…