CVE count tops 100,000 in 2026, driven by AI-assisted bug hunting
Disclosed vulnerabilities have more than doubled year on year, with AI tooling credited for the surge and the window before exploits appear shrinking to hours.
CRUNCHY TECH
EVERY PUBLISHED BRIEF
Browse Crunchy Tech’s latest security briefs, grouped by date.
PUBLISHED
Disclosed vulnerabilities have more than doubled year on year, with AI tooling credited for the surge and the window before exploits appear shrinking to hours.
A hack of email provider Arara hit the city's administrative platform, days after Osaka Prefecture disclosed its own breach amid a national surge in attacks.
Cloudflare reports a 30x rise in AI penetration testing traffic since October, with thousands of IPs running autonomous frameworks despite the original tool's takedown.
An international operation removed the Gootloader malware from more than 100 servers, part of a wider sweep that disrupted 235 servers and 257 domains.
Hackers stole driver's licence information via Uber's tech provider Fireworks AI, with state filings revealing thousands affected and the count expected to rise.
Daybreak Red Team and Defense Factory use AI agents to probe web apps and prepare fixes, though engineers still approve every patch.
The UK privacy watchdog has written to more than 200 senior health leaders, giving them until 27 November to personally confirm their safeguards against staff prying into patient files.
A new proof-of-concept exploit abuses the Windows Common Log File System and Defender to gain System privileges, with no sign of exploitation in the wild so far.
Attackers chained the patched PHP bugs with a Check Point privilege escalation flaw, hitting Israeli targets first before a second, non-destructive wave worldwide.
The new Humble Bundle site offers no way to set up 2FA, and users who had it are no longer asked for codes. The company has not explained why.
PUBLISHED
Microsoft confirms invisible zero-size images can stop documents printing, while a second bug breaks command-line activation tools on some Windows versions.
The airline cites cybersecurity risks and federal aviation rules as it moves to stop personal AI assistants booking flights or managing reservations.
A cyberattack at Fireworks AI, its second in as many months, let hackers steal billing data, datasets and some Notion AI chat transcripts.
Bucharest's intelligence agency says it contained a December intrusion by the FSB-linked Turla group, which used a new toolkit to spy on emails exchanged with NATO and EU institutions.
Attackers abused a flaw in a ScreenConnect extension to compromise dozens of organisations within 90 minutes, prompting ConnectWise to disable the affected extensions.
Billing records, datasets and partial model weights were accessed in two attacks, with Notion AI chat transcripts among the exposed customer data.
A 9.8-severity bug in Quay's key server could expose private images and secrets, and disabling anonymous access does not close the door.
PUBLISHED
A blanket ban takes effect immediately because the army has no central record of which units bought which drones.
Over 50 advisories cover critical TDX authentication bugs, dozens of AMD graphics flaws and a discontinued Intel tool users are told to uninstall.
PUBLISHED
Barracuda says a small number of organisations have been hit by exploits chaining a pre-auth bypass and command injection in its Email Security Gateway.
The Git project has shipped fixes for six flaws, including arbitrary code execution during recursive clones, and urges users to update bundled copies of Git too.
November's 42 advisories include four maximum-severity fixes, with researchers demonstrating working exploits for unauthenticated remote code execution in SAP kernel components.
Google DeepMind's AlphaProof Code formally verified Signal's PQXDH handshake and Triple Ratchet down to machine code, with humans reviewing just one percent of the codebase.
A researcher found a password embedded in the hotel giant's Android app that opened a Salesforce-hosted database holding names, emails and login tokens. Wyndham revoked the credential but won't say if customers will be notified.