Tag
All topics →Securing AI systems: the new attack surface across models, agents, pipelines, and the supply chain behind them.
40 stories, newest first.
Latest
The NIST AI RMF is voluntary, is not a checklist, and is being revised. What Govern, Map, Measure and Manage actually ask for, what the Generative AI Profile's 12 risks add, and a twelve-artefact minimum viable version for a SaaS company.
Read the article
Ten testable predictions for identity and AI security in 2027, from Know Your Agent rules to passkeys at 45% and EU wallet delays, each with a confidence level and a test.
LiteLLM's poisoned 1.107.0 release began with an admin key leaked in May and never revoked. It unlocked an agent platform holding the owner's unscoped GitHub token. What happened, and what to fix this week.
A free 26-question assessment scores your AI readiness across strategy, data, talent, use cases, culture, and governance. Here is how it scores, a sample run, and what to do at each tier.
AI made attacks cheap, but 2026's breaches still walked through old doors: exposed devices, default credentials and over-trusted agents. Here are six attack patterns and nine fixes, ordered by how fast they cut risk.
OpenAI has notified more than 100 organizations that its AI agents may have accessed their systems without authorization. The cause was not a smarter attacker. It was exposed credentials, legacy token endpoints, and agents nobody was watching. What happened and what to do this week.
Anthropic, OpenAI, Google, xAI and DeepSeek all retired or scheduled model retirements in 2026, and some old names quietly served new models. Here is what changed, why it matters, and how to use the AI Models lifecycle page to build a model-retirement calendar.
Microsoft's Defender team catalogued 31 companies across 14 industries embedding hidden instructions in "Summarize with AI" buttons to bias assistant recommendations. One sold web security software. After fifteen years in security, here is why AI visibility became a governance problem.
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.
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.
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.
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.
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.
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.
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.
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.
The AI observability market split into four segments (infra telemetry, dev tooling, runtime security, autonomous remediation) with a wave of 2025-2026 acquisitions behind it. None of them own whether an agent behaved correctly.
A leaked static key is a disaster; a five-minute token is mostly a shrug. Here is the credential lifecycle that gets AI agents from 24-hour tokens to ephemeral ones.
Shadow AI is an employee pasting into ChatGPT. A shadow agent holds real credentials and acts on its own. That is a different, and bigger, problem.
Mandiant ranked voice phishing the second most common initial infection vector of 2025. In the same window, thousands of businesses handed their phone lines to AI agents. Those two facts are related, and the security implications run in both directions.
AI made it cheap to fake a face, a voice, and a video. Here is my working map of the three problems authentication now has to solve, and the tools that actually hold up in 2026: verifying people, verifying agents, and verifying content.
Your AI security review passed and still missed the real attack surface. EchoLeak, over-permissioned agents, shadow AI: the AI-specific vectors most CISOs never test for, and the five moves that close them.
The Salesloft Drift breach hit 700+ companies with stolen OAuth tokens and never touched a password. Machine identities now outnumber humans 109 to 1, and AI-powered attackers are harvesting them at machine speed. Here is why human-shaped IAM cannot protect AI agents, and what to fix in 90 days.
Hugging Face detected the intrusion on July 16. OpenAI worked out five days later that the attacker was its own model, cheating on its own benchmark.
Agents can already prove who they are. What no standard has cleanly solved is passing scoped authority down a multi-hop chain across organizations. Here is the real state of agent identity in 2026, minus the blockchain hype.
Machine identities now outnumber humans by 45 to 1 or more, and every AI agent widens the gap. Here is what an identity orchestration layer is, in plain terms, and how to build one that governs humans, workloads, and agents from a single control plane.
Traditional vulnerability cycles take months from discovery to patch. AI compresses that to hours. CrowdStrike reports 29-minute breakout times and 89% more AI-augmented attacks year-over-year.
MCP concentrates every AI-agent-to-tool interaction through one interface, and that makes it a new attack surface. A breakdown of the vulnerabilities and how to secure them.
The leading non-human identity (NHI) management tools for 2026, compared by job: discovery and posture, secrets management, machine identity, and workload identity.
The leading CNAPP tools for 2026, compared by collection model and breadth: agentless-first platforms, broad full platforms, agent-led tools, and cloud-native specialists.
A poisoned LiteLLM package led to 4TB stolen from Mercor, the AI training startup serving Meta, OpenAI, and Anthropic. Class action lawsuits filed.
How to build a cybersecurity product roadmap that survives AI security, compliance deadlines, and threat-driven emergencies. A founder's four-lane framework.
Whether you're considering Grok AI for cybersecurity operations, customer engagement, content creation, or internal knowledge management, this resource
AI agents need to prove both their own identity and that a human authorized the action. Here is the three-token delegation architecture that answers both, and the security layers around it.
Discover how robots.txt has evolved from basic crawler control to a sophisticated security instrument.
Finding the right equilibrium between leveraging generative AI for cybersecurity and mitigating its potential drawbacks is crucial.
