02WRITEUPS

Notes from
the field.

Long-form explainers on AI security, identity, cryptography and risk. Read them in full here, then join the discussion on Medium.

AI · ADVERSARIAL MLLATESTWhy Every AI Security Professional Should Understand FGSM, BIM and PGD

How gradient-based attacks nudge a model's input just enough to flip its prediction, how the three methods differ in cost and strength, and why defenders should use them to test robustness.

JUL 2026 · 5 MINREAD

CRYPTOGRAPHY5 MINDiffie-Hellman: Establishing Secrets over Insecure Channels
How two parties agree on a shared secret in public and why that alone invites a man in the middle, followed by hands-on OpenSSL parameter audits, HMAC checks and cracking an unsalted MD5 hash with Hashcat.
IDENTITY · ACCESS5 MINUnderstanding Identity and Access Management (IAM)
Identification, authentication, authorisation and accountability explained, why home-made auth protocols fall to replay attacks, and how access control models and single sign-on compare.
RISK · THREAT MODELING5 MINThreat Modeling and Risk Assessment Frameworks
Threats, vulnerabilities and risk told apart, attack trees for mapping an adversary's path, and when MITRE ATT&CK, STRIDE, DREAD or PASTA is the right fit for a team.
RISK · GOVERNANCE5 MINNavigating Security Governance, Frameworks, and Compliance
How governance, risk and compliance fit together, the policy and standards hierarchy, GDPR penalty tiers, NIST SP 800-53 and 800-63B, and what ISO 27001 and SOC 2 each prove.
CRYPTOGRAPHY3 MINCracking the Code: A Practical Guide to Classical & Modern Cryptography
From Caesar shifts, transposition and frequency analysis to AES, RSA and the hybrid encryption behind HTTPS, with worked examples of why each classical method falls.
AI · SECURITY3 MINThe Future of Cybersecurity Has Changed: What Project Glasswing Means for the World
What Anthropic's Project Glasswing is, what its unreleased Claude Mythos Preview model reportedly found, and why AI now speeds up both attack and defence.
AI · ALIGNMENT6 MINUnderstanding AI Safety and Alignment: Why It Works, Why It Fails, and the Weak Foundation of Current Methods
Why base models are unsafe by default, how RLHF teaches refusals as learned probabilities rather than rules, why those safeguards break, and what realistic progress looks like.
AI · LLM SECURITY8 MINPrompt Security and Context Window Integrity in LLMs
What actually sits inside a model's context window, how prompt injection, prompt leakage and context poisoning abuse it, and the defences that help, with their honest limits.
AI · PERSPECTIVE2 MINAI as a Collaborator: Are You Adapting or Standing Still?
Why AI is better treated as a collaborator than a competitor, and why the people who adapt to it, rather than resist it, will shape how value is created.

Follow on MediumNew writeups land there first, with comments and claps.

Notes fromthe field.

Notes from
the field.