RESEARCH
Offensive research & field notes
Technical notes on AI and agent red teaming, sandbox containment, incomplete-fix variant analysis, and security due diligence.
2 October 2026
2 min read
AI Security
Testing containment and tenant isolation in AI agent sandboxes
When you give an autonomous agent code execution, browser tools, and short-lived credentials, your sandbox boundary is your primary security control. Here is how we test whether it holds.
Vulnerability Research
Why security patches fail: hunting incomplete-fix variants across source trees
When a vendor patches a vulnerability, they often fix the reported entrypoint while leaving sibling call-sites on the old guard pattern. How we systematically hunt and prove patch variants.
Governance
An EU AI Act readiness checklist for leadership
The EU AI Act is now enforcement reality, with obligations arriving on a schedule and penalties of up to 7% of global turnover. Here is what leadership actually needs to have in place.
Engineering
Your agent demos well. Production doesn't care.
The distance between an agent that impresses in a demo and one that runs unattended for months is not more AI. It is a specific list of engineering work, and this is the list.
CASE STUDY · Due Diligence
Deeper answers on every target
Why M&A security due diligence has to start outside the data room, combining external shadow-surface reconnaissance with targeted code, cloud, and corporate control-chain analysis.
CASE STUDY · Digital Assets
Following the money on chain
How a digital asset compliance team used AI agents to discover links between fraudulent crypto transactions across wallets, chains, and exchanges.
Security
Adversarial Review: The Missing Layer in AI Governance
One way to increase trust is to make a proposed action defend itself against a critic before it runs. Debate is a design pattern, not a guarantee of safe autonomy.
VANTAGEA RESEARCH NOTES
Offensive research, in your inbox
Technical write-ups on AI and agent exploitation, sandbox containment, variant hunting, and security engineering.