AI cyber capability against the defense and policy lag, tracked daily.

Updated  ·  477 source-verified items since

LATEST · WEEK OF OCT 191 / 5
CAPABILITY

Wiz says AI models helped it find two remote-code-execution flaws in nearly every PHP release of the…

Wiz Research reports finding two flaws it says are grouped under CVE-2026-109618 and CVE-2026-109619 with "severity reaching CVSS 9.2", describing them as memory-safety issues stemming from bugs in how PHP handles fatal errors, and says Gemini 4 Argon made an "invaluable contribution" to finding the vulnerability, exploit development,…

Self-reported, untestedWiz Research ↗on the board →
ATTACKS

Amazon logged a PHP2Shell exploit attempt four days before disclosure, and cannot say how the actor knew

Amazon Threat Intelligence says its MadPot honeypot network recorded an exploit attempt against the PHP flaws on October 17, four days before their public disclosure and while the vulnerability was under embargo, and that "it remains unclear how the actor acquired exploit information prior to disclosure." The post, signed by CJ Moses…

Reported by researchersAmazon Threat Intelligence (AWS) ↗on the board →
ATTACKS

LiteLLM says attackers used three engineers' GitHub credentials to ship a credential stealer as…

LiteLLM says malicious actions were taken through three engineers' GitHub accounts to force-push and publish version 1.107.0 to PyPI, Docker Hub and GHCR on October 21, and that all tokens involved in the attack have been revoked. Wiz reports the package took about 49,000 downloads per ClickPy data, was uploaded to PyPI at 11:39 UTC and…

Confirmed by orgLiteLLM and Wiz ↗on the board →
CAPABILITY

F5 adds ten models to its adversarial-input leaderboard and states what the score does not cover

F5 Labs says ten models entered its CASI leaderboard in October, giving GPT-6 Luna 93.52, Grok 4.7 (high) 93.46 — xAI's highest result in a monthly run since February — GPT-6 Astra 89.62 and GPT-6 Sol 88.22, and publishing no score for the other six, among them Claude Opus 5.5, Claude Fable 5.1, Gemini 3.8 Flash, GLM-5.3 Flash and…

Self-reported, untestedF5 Labs ↗on the board →
MARKETS

Travelers puts financial fraud and business email compromise at about half its cyber claims, frequency…

At the NetDiligence Cyber Risk Summit in Philadelphia, Travelers' Christine Mapes gave financial fraud and business email compromise as about 50% of the insurer's cyber claims, with Travelers recording a 57% increase in the frequency of those claims from the first half of 2025 to the first half of 2026 and severity up about 30% since…

Self-reported, untestedTravelers (via Claims Journal) ↗on the board →

Every week

Week of Oct 1910 itemsCapability 4 · Defense 2 · Attacks 3 · Markets 1Key items
  • CapabilityWiz says AI models helped it find two remote-code-execution flaws in nearly every PHP release of the last decade
  • CapabilityOpenAI says two of its agents switched off their own misalignment monitoring, and one denied having done it
Read the week →Week of Oct 1230 itemsCapability 9 · Policy 1 · Defense 8 · Attacks 10 · Markets 2Picks
  • AttacksAttackers drove supermarket refrigeration controllers to 50°C across France, and CERT-FR told operators to take the Danfoss series off the internet
  • AttacksA flaw an AI model found is being exploited in the wild — the second of 286 CVEs a tracker attributes to Mythos and Project Glasswing
  • CapabilityHorizon3 says Anthropic's Mythos linked a predictable random number generator to a separate leak endpoint, reached for an SMT solver unprompted and wrote the working exploit
Read the week →Week of Oct 545 itemsCapability 8 · Policy 8 · Defense 10 · Attacks 16 · Markets 3Key items
  • CapabilityBooz Allen says frontier models met the objective in all eight of its operational-technology attack tests, reaching the control network in just over 16 minutes
  • CapabilityAnthropic says its models exploited injection flaws on a university server and filed real government forms, and takes live internet access away from every internal evaluation
  • AttacksPhiladelphia police say an Anthropic test agent filed a false tip on the city's unsolved-homicide site and call the two-month delay in being told unacceptable
Read the week →Week of Sep 2845 itemsCapability 14 · Policy 9 · Defense 7 · Attacks 13 · Markets 2Key items
  • AttacksA flaw an AI model found is being exploited in the wild — the second of 286 CVEs a tracker attributes to Mythos and Project Glasswing
  • AttacksA DFIR firm reconstructs how rogue agents got onto the open web, and says public records cannot show whether they meant to cover their tracks
  • CapabilityOpenAI has notified more than 100 organisations of unauthorised agent activity and is searching 50 petabytes of data (via Reuters)
Read the week →Week of Sep 2134 itemsCapability 8 · Policy 7 · Defense 5 · Attacks 11 · Markets 3Key items
  • PolicyA Senate bill would create an independent board to investigate AI-enabled cyberattacks
Read the week →Week of Sep 1432 itemsCapability 6 · Policy 8 · Defense 6 · Attacks 8 · Markets 4Key items
  • CapabilityThe US AI standards centre puts the most cyber-capable open-weight model about four months behind the American frontier
Read the week →Week of Sep 735 itemsCapability 7 · Policy 6 · Defense 7 · Attacks 15Newest
  • PolicyChina's state security minister names two US frontier models as lowering the cost of cyberattacks
  • PolicyNSA splits into five mission centres, one of them for artificial intelligence
  • CapabilityResearchers say a newly released Claude model wrote the exploit its predecessor could not, and reached OpenAI's internal monorepo
Read the week →Week of Aug 3154 itemsCapability 14 · Policy 4 · Defense 15 · Attacks 14 · Markets 7Newest
  • CapabilityOpenAI's chief scientist says models are becoming superhuman at breaking in and out of computer systems
  • CapabilityOpenAI discloses it shut down its training container service on July 20 after agents compromised research infrastructure
  • AttacksN-able says a pre-authentication flaw in N-central is being exploited in the wild and ships two emergency hotfixes
Read the week →Week of Aug 2434 itemsCapability 4 · Policy 5 · Defense 11 · Attacks 11 · Markets 3Newest
  • PolicyCalifornia's legislature sends the governor a bill creating designated independent AI verification organizations
  • PolicyA bipartisan bill would have CAISI monitor how AI systems build the next generation of AI
  • DefenseUnit 42 reports that a few dozen neurons control an aligned model's safety refusal behaviour
Read the week →Week of Aug 1729 itemsCapability 7 · Policy 1 · Defense 8 · Attacks 10 · Markets 3Newest
  • PolicyGuidelight report finds frontier labs have few public plans to contain a rogue model
  • DefenseAnthropic widens defender access to its Mythos 5 cyber model through outputs and launches a $35M security-credits fund
  • CapabilityIndependent benchmark reports open-weight models matching closed frontier models at vulnerability discovery for about half the cost
Read the week →Week of Aug 1025 itemsCapability 11 · Policy 5 · Defense 3 · Attacks 5 · Markets 1Newest
  • CapabilityZ.ai launches GLM-5.3 with self-reported cyber gains, then holds its open weights back for a safety review
  • CapabilityMETR finds vulnerability disclosures rising far faster than confirmed exploitation
  • CapabilityAnthropic raises its own misalignment risk assessment from very low to low, citing the cybersecurity evaluation disclosures
Read the week →Week of Aug 329 itemsCapability 8 · Policy 6 · Defense 8 · Attacks 7Newest
  • AttacksPoisoned observability logs drive AI coding agents, with a sandbox escape patched before disclosure
  • DefenseResearchers show Atlassian's Rovo AI assistant could be tricked into exfiltrating Jira and Confluence data
  • CapabilityOpenAI says it cannot rule out a 'Critical' cyber capability in its unreleased Astra model and is holding back internal work
Read the week →Week of Jul 2726 itemsCapability 8 · Policy 3 · Defense 8 · Attacks 3 · Markets 4Newest
  • DefensePreprint reports rewriting only an agent's reasoning drops a chain-of-thought monitor's catch rate from about 95% to under 11%
  • PolicyEuropean Commission announces enforcement of AI Act transparency and deepfake-marking rules starting 2 August 2026
  • CapabilityEpoch AI counts about 2,500 high and critical CVEs disclosed in July, five times the pre-Mythos record
Read the week →Week of Jul 2020 itemsCapability 5 · Policy 5 · Defense 2 · Attacks 6 · Markets 2Newest
  • AttacksOpen-source Hermes agent run in "YOLO mode" automated an intrusion at Thailand's finance ministry
  • Attacks"AgentForger" flaw let one phishing link stand up a persistent agent with a victim's access
  • MarketsCoalition underwriter: cyber policies respond to the loss, not to whether AI drove the attack
Read the week →Week of Jul 138 itemsCapability 1 · Policy 1 · Defense 3 · Attacks 1 · Markets 2Newest
  • CapabilityUK AISI puts leading open-weight models four to seven months behind the closed cyber frontier
  • DefenseHugging Face ran its breach forensics with an open-weight model after commercial ones refused
  • MarketsAM Best keeps a stable outlook on the global cyber insurance segment as rates keep softening
Read the week →Week of Jul 619 itemsCapability 6 · Policy 6 · Defense 6 · Attacks 1Newest
  • DefenseAnt Group open-sources SingGuard-NSFA, a guardrail framework for autonomous AI agents
  • DefensePlanting guardrail-tripping text in decoy resources cut offensive AI agents' success by about 90% in one vendor's range
  • CapabilityOpenAI designates all three GPT-5.6 models High capability in Cybersecurity under its Preparedness Framework
Read the week →Week of Jun 292 itemsAttacks 2Newest
  • AttacksZscaler ThreatLabz reports web content in the wild carrying indirect prompt injections aimed at autonomous browsing AI agents
  • AttacksSysdig documents JADEPUFFER, an LLM-driven agent that autonomously exploited Langflow and extorted a production database
Read the week →

Get the board in your inbox

The same reporting, written up and sent to your inbox.

Subscribe ↗