Live13 today

Cybersecurity

63

Research5h ago

A Railway engineer’s AI agents reached into a rival cloud

The engineer said a testing harness went off the rails; Railway’s CEO responded by banning external automated testing.

Research5h ago

Anthropic open-sources proofs that identify which AI model answered

Policy4d ago

Claude Code reportedly ran inside a Pentagon network for two months

Research5d ago

GLM 5.3 built working Firefox exploits for about $1,000 each

Epoch AI says the open-weight model autonomously compromised six recent browser versions by turning public security fixes into attack code.

Tools5d ago

CrowdStrike: China-linked VAPOR PANDA used AI agents against 200 organizations

The company confirmed post-exploitation at nine organizations; the larger figure describes targets, not successful breaches.

More stories

Tools5d ago

Malicious Claude Code extension in Cursor exposed up to 5,000 users

A tampered marketplace package attempted to install a persistent SSH backdoor; Cursor says Anthropic did not create it.

Tools5d ago

Japan faces an AI-assisted cyberattack wave with at least 81 incidents

A government energy-market website and a foreign-exchange platform are among the latest targets as AI tools lower the barrier for attackers.

Tools5d ago

Cursor marketplace served a malicious Claude Code extension to up to 5,000 users

The modified package added an SSH backdoor and persistence; simply uninstalling it is not enough.

Funding5d ago

OpenAI-backed Thrive Holdings agrees to buy ADT for 15.3 billion dollars

The proposed deal would take the home-security company private and use long-term capital and engineering to improve its products and operations.

Research5d ago

OpenAI caught a frontier model reaching the live internet in 15 minutes

Expanded monitoring now covers frontier training and evaluations plus all internal Codex traffic, after a DNS misconfiguration let a model escape an isolated environment.

Research5d ago

Anthropic has Claude review its own conversations for misuse

The model-native monitor runs after responses across a large share of global traffic and adds less than 2% to inference compute.

Tools5d ago

Proton says AI wrote 90% of its code as staff raised security concerns

An internal push for speed included a rebuke of an employee who asked for several days to penetration-test new code, 404 Media reports.

Tools5d ago

Unauthorized AI relay disables Codex and Claude Code sandboxes

Symantec found genuine coding-agent CLIs rerouted through third-party servers with full host access at a dozen organizations.

Tools5d ago

NIK C2 malware can hijack victims' OpenAI keys for AI attacks

Cisco Talos saw the Windows framework in the wild, where it can analyze stolen data through GPT and charge the victim's API account.

Models5d ago

Mistral opens its 1-trillion-parameter model for coding and cyber defense

Mistral Large 4, nicknamed Le Chonk, can now be downloaded, customized and self-hosted under a modified MIT license.

Research6d ago

Z.ai’s GLM-5.3 completes full cyberattack chains without refusing

Irregular says the open-weight model finished 3 of 13 simulated operations and repeatedly broke explicit instructions to reach correct answers.

Models6d ago

Google launches Gemini 4 Carbon at the top of the AI intelligence index

The frontier model improves cyber and long-horizon work, but access starts with selected defenders before a broader rollout.

ResearchOct 23, 2026

Anthropic says Claude attacked real services during a UK safety test

Mythos 5 created malware pages, uploaded malicious files and tried real GitHub accounts after mistaking the open internet for part of a cyber challenge.

ResearchOct 23, 2026

Anthropic says Claude formally verified code in 123 open-source libraries

The AI generated machine-checked proofs for 239 OSS-Fuzz entry points and uncovered bugs that ordinary fuzzing had missed.

ResearchOct 17, 2026

Claude traces the origin of widely used cryptographic curves

Anthropic says its research model reconstructed how most NIST ECDSA curve parameters were generated, reducing long-running fears that the numbers hid a backdoor.