Security research, product updates, and supply chain threat intelligence.
What SURADAR records for an EU AI Act programme, which articles that evidence maps to, and which obligations stay with the provider and deployer. Glyphzero, Inc. does not certify compliance.
Read more →Varonis found a Dialogflow CX flaw where one edit permission on one agent could read conversations, steal credentials, and impersonate the bot. The failure was plumbing, not prompts.
Read more →A live sandbox of an enterprise CFO agentic workflow: one CFO agent, six negotiation agents, payment execution, FP&A. We deployed malicious agents inside the sandbox to simulate real-world prompt injection and rogue delegation. Without SURADAR, the rogue agent moved $246K in unauthorized terms. With SURADAR, every attack vector was blocked, every action was logged, and zero unauthorized dollars left the system.
Read more →Fragmented virus payloads, Denial of Wallet compute exhaustion, agent supply chain attacks, cross-agent privilege escalation, 22-second breaches, self-modifying agentic malware, and the consent revocation race. The threats that will define 2026.
Read more →Cedar is a solid policy engine, but it has a blind spot: it trusts whatever identity claim it receives. AI agents are exploiting this gap at scale. Here's what's missing and what needs to change.
Read more →Google DeepMind's new taxonomy maps six ways attackers hijack autonomous AI agents. Here's what it means, and how per-request authentication changes the equation.
Read more →Six months of patience, two developer laptops, one weaponized VS Code folder, and 31 transactions. The Drift Protocol hack is the most technically sophisticated social engineering operation ever executed against DeFi.
Read more →The full post-mortem of the Mercor breach: how a cascading supply chain attack through Trivy and LiteLLM gave Lapsus$ full VPN access to an AI hiring platform trusted by OpenAI and Anthropic.
Read more →The full post-mortem of the March 2026 TeamPCP supply chain attack on LiteLLM: how attackers weaponized a trusted vulnerability scanner to hijack 95 million monthly downloads and harvest every credential on infected machines.
Read more →The full post-mortem of the March 2026 CodeWall breach of McKinsey's Lilli: SQL injection in JSON keys, 22 naked endpoints, prompt layer compromise.
Read more →An archived introduction to Code Corgi and invisible threats in source code.
Read more →Vibe-coded applications skip authentication, hardcode secrets, and ship without security headers. Review these failure modes before release.
Read more →How attackers embed malicious instructions in calendar invites, emails, and documents to hijack AI assistants, and the boundaries teams can enforce.
Read more →How a single malicious Google Calendar invite gave attackers control of Gemini's AI assistant, exfiltrated emails and location data, and opened smart home windows, without the victim clicking anything.
Read more →