Writing
How attacks work, what AI is doing to them, and what I have learned running a security company since 2006. Pieces marked with a site name open on that company’s blog.
2026
If an attacker obtains valid credentials, what stops them executing financial transactions? Login cannot be the whole security model.
The concern isn't any single AI-powered attack. It is what happens when attacking one more organisation becomes cheap.
Attackers are coming through fintech integrations, NBFCs, vendors, exposed APIs and stolen credentials. Notes from The Morning Brief.
- Inside B-52’s fully autonomous run on all 104 XBOW challengeson b52security.com
How the run was set up, and how fresh flags, captured evidence and a separate referee decided each solve.
- How to evaluate an agentic pentest on your own applicationon b52security.com
Use your own workflows, synthetic accounts and scope boundaries, and reproduce the findings independently.
- The eMudhra CA key compromise: why your certificate authority belongs in the vendor risk registeron shadowmap.com
Eight certificate authority keys were trivially weak; one was the number 4. Which systems may still trust them, and what to ask your CA.
An exposed attacker server showed an always-on discovery operation. Attackers think in paths, and cadence matters more than most admit.
- AI pentest is two jobs pointing in opposite directionson b52security.com
One name covers two opposite jobs: autonomy that runs the test, and a test whose target is the AI.
- A gate that fails every build is a gate somebody turns offon b52security.com
Security testing in CI/CD needs a threshold you set, and each severity setting shapes a team differently.
- The report is finished. That is exactly when it gets checkedon b52security.com
PTES and NIST SP 800-115 both stop at reporting. The case for a quality gate that runs on the finished report.
- Testing the inventory discovery foundon b52security.com
Discovery returns hosts your asset register does not have. What happens to that tail, and who signs off the scope.
- What autonomous persistence means under authorisationon b52security.com
Holding a foothold over time in automated red teaming: under whose authority, and bounded by what.
The news isn't really one AI model. AI is compressing security audits from weeks to hours, and the same is coming to attackers.
2024
Dark Web monitoring vendors, ShadowMap included, buy data from threat actors. Are customers funding the people who target them?
2020
- What the Zhenhua Data investigation revealedon shadowmap.com
An investigation of the Chinese intelligence firm Zhenhua Data, based on data collected by ShadowMap.
2015
The 2008 meeting that sent me back to rethink Security Brigade's USP, and how its second question led us to build Lemon.
2012
Three times in a few weeks my contact details turned up where I never gave them, and what that says about privacy in India.
2010
Handing my phone in for repair made me wonder why so few businesses add one small security step to their everyday processes.
2009
A bank asked me for a blank signed cheque. What a process like that teaches customers to accept.