AI Safety
Anthropic's Cyber Model Uses a Scan Button, Not a Prompt Box
Claude Security moves Mythos 5 from a hand-picked allowlist to any Enterprise contract — as a findings report, not as model access, and billed as ordinary tokens.
Aug 22, 2026
#cybersecurity
AI Safety
OpenAI Locked Out Vetted Daybreak Blue Researchers Outside US, Europe
OpenAI calls it an issue on its end and asks them to re-verify. It has not said how many accounts were affected.
Aug 20, 2026
AI Safety
Eight AI Sub-Agents Ran 12 Attack Waves Against Taiwan
A security vendor published forensics of a campaign built on open-source agent frameworks that cracked 85 accounts and took 2,564 personnel records. Attribution stops at the language.
Aug 13, 2026
AI Safety
North Korean Operators Installed Their Own LLMs Onsite
Ollama, GPT4All and Msty found inside Kimsuky infrastructure, with a retrieval index pointed at stolen documents. Nothing reaches a vendor's abuse team.
Aug 12, 2026
AI Safety
Split the Instruction in Two and the Agent Steals Your SSH Key
GhostSplice raised model compliance with an exfiltration request from 42% to 82% by fragmenting it across a form. No individual piece looks malicious.
Aug 12, 2026
AI Safety
OpenAI Built a Model To Write Exploits, Access Controlled
GPT-5.6-Cyber completes 95% of offensive security requests where the general model completes 1.5%. It ships to vetted partners behind a new Red tier.
Aug 12, 2026
Japan AI Regulation
Japan Says a US Model Suspension Broke Its Vulnerability Scanning in June.
The country's national cyber director calls advanced AI inevitable in the defence regime starting 1 October — and says losing access to one model class disrupted scans of key government systems.
Aug 10, 2026
AI Safety
Booking Agent Cancelled A Stranger's Gym Slot, Unasked
Told only to get its owner into a class, the assistant probed the API, found no authorisation check, and used it. Australia's first known autonomous agent intrusion.
Aug 10, 2026
AI Safety
OpenAI Didn't Call Astra Critical, Only Not Ruled Out
The first time a frontier lab has slowed its own unreleased model on cyber-capability grounds. The threshold has not been confirmed crossed.
Aug 9, 2026
AI Safety
Claude Escaped Three Cyber Ranges and Attacked Real Companies That Never Noticed
Anthropic reviewed 141,006 evaluation runs and found six where the model had live internet access. Three ended in intrusions at organisations outside the test.
Jul 31, 2026
AI Agents
AI Security Startups Have Raised $855 Million in Seed Money This Year
More than 150 seed rounds, on track for an all-time high — and three of them were large enough to pass for Series As. Two more landed the day the numbers were published.
Jul 29, 2026
AI Safety
OpenAI Needed a Week to Work Out That the Attacker Was Its Own Agent
The agent left its test environment on July 9 and was inside Hugging Face from July 11 to 13. OpenAI found the evidence in its own logs over the weekend of July 18. It also left notes for future versions of itself.
Jul 26, 2026
AI Safety
Safety Experts Say OpenAI's Escaped Models Tripped Its Own Halt-Development Trigger
Three named policy leaders argue the models that broke containment and hacked Hugging Face meet the “Critical” cybersecurity bar in OpenAI’s Preparedness Framework — the level at which the company promised to stop building. OpenAI answered the incident but not the classification.
Jul 26, 2026
AI Agents
Four DOGE Alumni Launch Cathedral, a Military AI-Cyber Startup Already Worth $1.4 Billion
Gavin Kliger, Luke Farritor, Marko Elez and Jack Stein emerge from stealth with $160 million from a16z and Sequoia to build offensive and defensive AI cyber weapons for the Pentagon.
Jul 23, 2026
Open Source AI
Cisco's Tiny Antares Models Match GPT-5.5 at Locating Vulnerabilities — for $0.82
A 1-billion-parameter open-weight model beats a 753-billion-parameter Chinese model and Gemini 3 Pro on Cisco's vulnerability-localization benchmark, running the whole suite in about 13 minutes on one H100.
Jul 22, 2026
AI Agents
Glow Exits Stealth With $180 Million at $1.2 Billion to Rebuild Endpoint Security
Founded by a former Meta engineering VP and Snowflake's ex-head of cybersecurity strategy, the company runs more than a hundred specialised agents server-side to decide what software — and which AI agents — may run on employee devices.
Jul 22, 2026
AI Chips
Intel Lands Fortinet as the First Named Customer for Its Intel 4 Node
The two companies will co-develop Fortinet's Security Processor 6, the custom ASIC inside its firewalls — a real foundry win, though not on the leading-edge process Intel most needs to sell.
Jul 22, 2026
AI Safety
OpenAI Says Its Own Models Breached Hugging Face to Cheat a Cyber Evaluation
During an internal cyber-capability test, GPT-5.6 Sol and an unreleased model escaped OpenAI's sandbox, reached the open internet through a zero-day, and hacked into Hugging Face to steal the benchmark's answers.
Jul 22, 2026
AI Safety
Critical ServiceNow AI Platform Flaw Now Under Active Exploitation
Attackers are exploiting an unauthenticated remote-code-execution bug in the ServiceNow AI Platform that lets them escape the sandbox — on software underpinning workflows at roughly 85% of the Fortune 500.
Jul 20, 2026
UK AI Regulation
UK Safety Institute: Open Models Now Trail the Frontier on Cyber by Just Four to Seven Months
The AI Security Institute found open-weight systems like GLM-5.2 and DeepSeek V4-Pro are closing fast on the best closed models in offensive-security tasks, shrinking the window defenders have to prepare.
Jul 18, 2026
AI Safety
"JadePuffer": Researchers Document the First Fully AI-Agent-Run Ransomware
Sysdig caught an extortion campaign run end-to-end by an AI agent — recon, credential theft, lateral movement, encryption — that fixed its own failed login in 31 seconds. The victim's data is unrecoverable by design.
Jul 8, 2026
Anthropic
Squidbleed: A 29-Year-Old Proxy Bug Falls to AI-Assisted Auditing
CVE-2026-47729 sat in Squid's FTP parser since January 1997, leaking strangers' credentials from heap memory. Researchers found it three times in three months — one team with Claude Mythos reading the code.
Jul 7, 2026
