Notes
Technical write-ups. No product pitch, just the work.
2026-10-06 · Application security
Two critical unauthenticated RCE flaws in Next.js, an AVIF decoding bug and a Windows-only path traversal issue, show three blind spots automated scanners cannot cover.
2026-09-29 · AI red teaming
700 OpenAI agents recognized an attack on Hugging Face was out of scope and joined anyway. What that means for the difference between a model safeguard and an authorization boundary.
2026-09-23 · SaaS security
Socket traced 18 malicious npm packages that impersonated Alibaba tools and assembled a RAT across a dependency tree. What SaaS CI pipelines still miss.
2026-09-16 · AI red teaming
Oasis Security found a Claude Desktop bug that submitted a prompt from a claude:// link with no Send click. Why auto-submit deep links belong in AI red teaming.
2026-09-09 · Application security
CVE-2026-19478 is a CVSS 9.4 GraphQL authorization flaw, not RCE. Why unauthenticated writes on public objects still slip past scanners, and what to test on your own schema.
2026-09-02 · AI red teaming
Tenet Security showed at DEF CON how poisoned Cloudflare, Datadog, and Sentry data can make a trusted AI agent abuse its own access. What to map before you add another on-call agent.
2026-08-24 · Compliance
The CRA's reporting obligations become enforceable September 11, 2026. Most SaaS companies aren't sure if they're in scope. Here's how to actually check, and what to do if you are.
2026-08-21 · SaaS security
A compromised maintainer account turned a ten-year-old Rust crate with 245 million downloads into a build-time malware dropper. What the arrayref incident means for CI/CD pipeline security.
2026-08-14 · AI red teaming
What the OpenClaw prompt injection and agent phishing attacks reveal about AI agent security risks, and the controls SaaS teams need before shipping.
2026-07-28 · AI red teaming
Microsoft launched MAI-Cyber-1-Flash and an agentic platform called Perception. Here is what the arrival of a third major AI security model means for SaaS teams and authorization testing timelines.
2026-07-23 · Company
A short note on the problems we want to help European teams fix first.
2026-07-18 · AI red teaming
Yellow teams build the harnesses that make AI red teaming actually work, instead of just generating noise. What that means for SaaS teams evaluating AI security testing and build-versus-buy decisions.
2026-07-14 · AI red teaming
Orphaned AI agents keep production access after their creators leave, and traditional audits miss them. How SaaS teams inventory agents, scope permissions, and test for prompt injection before attackers do.
2026-07-09 · Compliance
A buyer-led framing of market expectations, not a standard-by-standard law lesson.
2026-07-07 · Application security
CVE-2026-8451 was exploited within 24 hours of WatchTowr Labs publishing a PoC. Why the patch-deploy gap makes reactive security fail for authentication infrastructure, and what proactive testing finds first.
2026-07-04 · SaaS security
Apple just proved the AI exploit patching cycle is real. Why annual pentests are compliance theater, and what testing cadence actually keeps up with AI-accelerated attackers.
2026-06-25 · Application security
OWASP, abuse cases, and how to pair automated signals with a focused manual review.
2026-06-11 · SaaS security
What to prioritise in the 12-24 month window: identity, data, and vendor risk without boiling the ocean.
2026-06-03 · Pricing
From pricing signals to what actually moves the number for B2B SaaS.
2026-05-27 · AI red teaming
Most SaaS companies fund AI agent identity security from the wrong budget. Here is what it actually costs and how to plan for it.
2026-05-22 · AI red teaming
What the Claw Chain vulnerabilities in OpenClaw reveal about AI agent attack surfaces, and what SaaS teams should be testing before they ship.
2026-05-14 · Pentest process
Deliverables, common coverage, and the difference between tiers in plain language.
2026-04-30 · Pentest process
Scoping, calendar time, and what the typical windows look like for SaaS teams.
2026-04-29 · Compliance
What auditors expect, when a pentest is in scope, and what you can de-risk without a full test.
2026-04-28 · Application security
CVE-2025-29927 is patched. Three specific Next.js auth patterns it exposed are not. How to audit your own app for each one.