Writing
Long-form essays on identity, AI security, CIAM, Generative Engine Optimization, and the practice of building software. 707 pieces, newest first.
Latest essay
Vendor log retention ends months before the average breach is discovered. The funnel, security, and silent-failure metrics every CIAM deployment should have, and the seven-panel dashboard to start with.
Read the article
Model providers solved the machine half of agent authentication in 2026. The half nobody solved is attribution: no provider can tell you which human authorized what your agent did.
Referral and inbound carry the lowest CAC because the buyer arrives already educated. But the cheapest channel is not automatically the right one: deal size and whether demand already exists decide which channels can work at all.
Zero data retention is not a setting, it is a contract term with a scope. The gap between what founders think ZDR covers and what it contractually covers is wide enough to fail an enterprise review.
ISO 42001 went from obscure standard to procurement line item in about eighteen months. It does not certify that your model is safe, and the certificate scope is where buyers consistently misread it.
Security practitioners are professionally trained to distrust claims, which makes them the hardest audience in B2B for conventional content marketing. The way through is not better persuasion. It is content specific enough to be checked.
Most founders think the EU AI Act is the model provider's problem. If your output is used in the EU, you carry obligations of your own, and the enforcement machinery went live in August 2026.
More than half of BIMI records are broken and nobody got an error. Silent failure is a whole category of security control, and most of yours are in it.
Most teams treat LLM output as deterministic. We measured 180,000 responses across four AI engines for 18 months and found the opposite: visibility drops 30% overnight, and the engines agree on 11%.
The handoff was never a document problem. What AI actually changes about moving a customer from sales to support, and why the same work makes AI engines understand your product.
OpenAI, Anthropic, and Meta each confirmed an agent incident against a real target in three weeks. The labs are right that it was a test. The capability is not.
Email, AI agents, media files, and build artifacts all got cryptographic verification in 2026. Same architecture, same failure modes, same mistake.
More than 20 Pass-the-Passkey techniques abuse logging, session handling, and synced-key recovery. None break FIDO2. Do not freeze the rollout.
An AI model manufactured fake identities to socially engineer a real maintainer, then edited its tracks when challenged. The request failed. The threat model should not.
Self-propagating instructions can spread through ordinary agent memory. Isolated agents with conflicting goals wrote malware to sabotage each other.
CrowdStrike clocks 29-minute breakout times and an 89% surge in AI-augmented attacks. Here is the five-phase framework I use with CISOs to close that gap, with budget splits and a board script.
Meta's settlement with 51 attorneys general commits at least $12.1 billion, but the money is the survivable part. Every teen safety term in the deal depends on knowing who is under 18, and nobody has solved that yet.
MCP's July 2026 spec rewrite went stateless and made Client ID Metadata Documents the standard, not audience-bound tokens, which have been mandatory since mid-2025. What actually changed since December 2025, and the checklist that replaces the old one.
Most cybersecurity vendors sell to the CISO and lose the deal to a security engineer they never spoke to. Security purchases are committee decisions with an asymmetric structure: many people can kill a deal, few can approve one. Here is how it actually works.
Paid AI placement sits beside the answer. Organic citation sits inside it. Buyers read them differently, and that asymmetry should drive the split.
CVE, CWE, CAPEC, ATT&CK, CVSS, KEV, ATLAS. The security taxonomy acronyms get used interchangeably and they should not be. Here is what each one actually does, how they chain together, and why they matter more for your content strategy than most security marketers realize.
GrackerAI switched enterprise SSO from WorkOS to SSOJet, cut the annual bill by roughly $5,000, and gained the custom authentication flexibility an AI platform needs.
Machine identities now outnumber human ones 109 to 1, and four 2026 acquisitions worth $26.6B prove vaults can't keep up. What ephemeral secrets and workload identity replace them with.
Ninety days is enough to fix crawler access, baseline measurement, and restructure your ten highest-value pages. Week by week.
Engineering teams resist AI initiatives over career anxiety and loss of control, not technical doubts. What actually worked leading teams through this at LoginRadius and GrackerAI.
Most companies ask how to help people get better at using AI. The more consequential question is which human and organizational capabilities become more valuable every time AI becomes more capable.
Most teams shipped AI agents without ever bounding what they can reach. The fix is a 1990s forward proxy, and an allowlist alone will not save you.
A decision framework for enterprise passkeys: when device-bound hardware keys beat synced passkeys, mapped to user risk, device context, compliance, and total cost. Includes the three failure patterns that surface only after rollout.
MIT found 95% of generative AI pilots produce no measurable return. After running AI agents in production for hundreds of B2B SaaS customers, here are the three warning signs I wish I'd caught earlier.
CPU cache is the fast memory between the cores and main memory. What L1, L2, and L3 each do, why cache lines are 64 bytes, and when more cache actually makes a processor faster.
WebAuthn Level 3 was proposed for W3C Recommendation in July 2026. Encryption key derivation, cross-domain credentials, and automatic list syncing are now first-class. Here is what changed.
