The landing site for Trackr Services
Varonis’s RovoBlast turned an Atlassian URL parameter into a trusted instruction, and Rovo’s ResearchAgent went and collected Jira, Confluence, and SharePoint data under the clicking user’s own identity. Atlassian fixed that link path in July. The identity and audit gap underneath it is still open.
Lazarus used an actively exploited afd.sys use-after-free (CVE-2026-68820) to deploy FudModule v3.1 on Windows 11 26100 and 26200. The rootkit’s first act is killing the crash dump path, which changes the order you triage a suspected host.
Metabase’s CVE-2026-72898 patch closes an unauthenticated SQL injection in the password-reset endpoint, but it does not remove the admin accounts, API keys, or rotated connected-database credentials an attacker created before you upgraded. Here is what the retro-hunt actually looks like in ingress logs and the Metabase application database.
Four CVEs across AWS, Google, and Vercel agent runtimes converge on one authorization failure: tool execution could proceed with no trustworthy binding to a model-authorized event. Here is what the audit record shows, what the detection actually looks like in CloudWatch and Splunk, and why the Strands branch that started it all is still open.
The SonicWall SMA 1000 exploit chain gets from unauthenticated HTTP to root through a loopback service and a control-service password derived from the appliance’s DMI product_uuid. Rapid7 and Resecurity assess INC Ransomware as the dominant actor now weaponizing it, and the loot — session databases and TOTP seeds — is why patching alone does not end the incident.
Hugging Face’s July 27 timeline of the autonomous agent intrusion reads as a sequence of identity operations: node impersonation via IMDS, TokenRequest minting, an over-scoped CSI ClusterRole spent on a privileged pod, and forged tokens from a stolen EdDSA signing key. The lessons sit in key custody and least privilege, plus one correction worth stating plainly: audience binding does not contain a compromised signer.
A working Windows exploit dropped roughly half an hour after July’s Patch Tuesday, with no CVE and no Microsoft advisory. LegacyHive edits a registry hive while it is unmounted, so live registry auditing never sees the write, and the payoff is another account’s hive mounted into a session the attacker controls. Here is what actually fires, what the first published rule got wrong, and which controls cut the class.
An indicator-driven study built a corpus from 1.2 billion Common Crawl URLs plus indicator-matched Censys and Shodan snapshots, and 51.2% of its 15,387 validated injections sat in custom HTTP response headers rather than page content. Here is what that number does and does not measure, why header presence is a serialization problem rather than a protocol problem, and why the representation your pipeline hands the model matters more than the header namespace.
A PRC-nexus actor exfiltrated years of medical and defense research by creating a single domain-level Gmail content compliance rule that silently BCC’d matching mail to an attacker Gmail account. It produced no endpoint telemetry and no user-visible forwarding setting, and the admin audit event family that authoritatively records it is mid-migration, with the legacy events retiring after August 2026.
A model closed an open question a 2025 cryptanalysis paper had left conditional, halving the effective keysize of an undeployed PQC candidate for about $100,000 and 60 hours of machine time. Nothing you run changed; the cost curve for producing novel cryptanalysis did.
