Latest posts
-
AI penetration testing: ThreatBook launches RedAgents
ThreatBook has launched RedAgents, an AI penetration testing aggregator that runs multiple agents at once. We explain the thesis, the dual-use risk and what remains unverified.
-
Zero-day vulnerabilities: 98 found at Pwn2Own Ireland
Pwn2Own Ireland paid out $1.26m for 98 zero-day vulnerabilities, with AI infrastructure and coding agents now on the target list. We examine the trust-boundary failures behind them.
-
AI-led cybersecurity: AAA Technologies pivots
AAA Technologies is repositioning as an AI-led cybersecurity company with four new services and a 100-to-1,000 workforce target. We look at the pivot, the DPDP deadline and the cost data behind it.
-
Decision models: AutoTrust tops Hugging Face trending
AutoTrust’s JEV-27B-VL has topped Hugging Face’s trending list. We look at decision models — models that return calibrated probabilities, not text — and what the milestone does and doesn’t show.
-
DDoS scrubbing: Malaysia’s new in-country defence
Zetvest has become Nexusguard’s authorised MSSP for Malaysia, rolling out a national DDoS scrubbing centre. We look at why in-country mitigation is a sovereignty play — and what 200Gbps can and cannot do.
-
AI agent security: Cequence takes on Meta’s Muse bots
Meta’s Muse is a customer and a bot at once. We look at the AI agent security problem behind Cequence’s new Agent Trust — and why identity alone is no longer enough.
-
Indirect Prompt Injection: Why a Guardrail That Fires After the Attack Has Already Failed
Salt Labs showed one email could hijack the Manus agent, run code, and reach connected accounts — and the guardrail that detected it fired too late to stop it.
-
Agentic Offensive Security: Palo Alto Networks Resells the AI Models Too Dangerous to Ship
Palo Alto Networks is renting out the AI models too dangerous to ship — via a subscription offensive-security service built on Anthropic’s Mythos and OpenAI’s GPT-5.6-Cyber.
-
AI Incident Response in India: Better Than the World, Still Not Good Enough
India tests AI incident response at 2.6 times the global rate — and still leaves half its enterprises unrehearsed. The survey numbers, decoded.
-
Proofpoint Partner Awards 2026 Celebrate a Channel Powering an AI-Security Pivot
Sixteen trophies in San Diego — and the refinancing, Microsoft bundling and AI-agent race that the ceremony never mentions.