Archive
Format: Security brief
-
REVIEW
LangChain Core serialization flaw could turn user data into trusted objects
CVE-2025-68664 affected serialization and loading paths; the advisory lists fixes in the 0.3 and 1.x branches.
-
REVIEW
Mixpanel incident exposed limited OpenAI user analytics, not chat contents
OpenAI’s updated notice distinguishes identifying analytics from account credentials, API content and conversation data.
-
REVIEW
ShadowMQ research traced unsafe deserialization across AI inference projects
Oligo’s findings concerned exposed messaging interfaces and Python deserialization, with different advisories for individual projects.
-
REVIEW
Apple credited Big Sleep for five WebKit findings in November 2025 updates
The affected component was WebKit; Google’s AI research agent was credited as a discovery tool.
-
REVIEW
OpenShift AI privilege-escalation flaw required an authenticated account
CVE-2025-10725 could turn low-privilege access into cluster administration; Red Hat rates the issue Important.
-
REVIEW
ForcedLeak research showed an Agentforce prompt-injection path through lead forms
Noma’s 2025 disclosure connected attacker-controlled form content with CRM access and an outbound trust weakness.
-
REVIEW
Cursor repository-opening risk involved automatic tasks and workspace trust
September 2025 research described how a malicious repository could exploit permissive task-execution defaults.
-
REVIEW
Nx’s s1ngularity postmortem traces a package compromise to stolen publishing access
The August 2025 supply-chain attack used malicious package scripts to search developer machines for secrets.
-
REVIEW
Stolen Drift OAuth tokens enabled Salesforce data theft in August 2025
Google’s investigation tied the campaign to compromised integration tokens, highlighting the reach of third-party SaaS access.
-
REVIEW
Cursor MCPoison flaw let changed tool configurations retain earlier approval
Check Point’s 2025 disclosure shows why tool approval must cover configuration changes, not just a trusted name.
-
Amazon Q’s 2025 extension incident highlights the release trust boundary
The reported incident involved malicious instructions in a software release. Payload intent is different from demonstrated damage.
-
Base44 disclosure shows how shared authentication flaws can affect private apps
Wiz reported an account-creation path that bypassed private-application access controls; the vendor fixed it after disclosure.