AOTP
Private AI→Proof
Privacy. AI acceleration. Human control.
Inspect →Proof
The capability is built, not borrowed.
When standard tooling falls short, we build the capability to turn uncertainty into proof.
Selected engineering and research
Security Engagement Platform
Engagement→Standard
One operating standard for authorized, isolated, evidence-ready delivery.
Inspect →OSMAP
Browser→Mail
Secure mail access for high-consequence systems.
Inspect →Browser AI Security
Page state→Evidence
Repeatable evidence for browser-AI attack paths and trust failures.
Inspect →Why we build
Innovation matters when it changes the outcome.
Black Bag Security writes code when it lets us test deeper, protect client information better, make evidence more defensible, or turn difficult work into a repeatable capability.
Engineering with a purpose
◎
Test deeperReach the security question commodity tooling cannot answer.
▰
Protect betterReduce unnecessary exposure of client and campaign information.
□
Prove clearlyMake evidence reproducible, reviewable, and tied to the claim.
⚙
Repeat reliablyTurn hard-earned technical work into durable operating capability.
Offensive security innovation
AI can accelerate testing.
Authority stays human.
AOTP and the Security Engagement Platform are built around two problems Black Bag Security cares about deeply: keeping sensitive engagement context under control, and keeping authority, evidence, and reporting connected to the human operator.
Private capability. Controlled delivery.
◉
Local intelligenceCampaign reasoning can stay close to the work rather than becoming another external data path.
▣
Human authorityScope, execution, stop conditions, material judgment, and findings remain governed by people.
□
Evidence-backedPotential findings move through proof, replay, validation, and review.
✓
Repeatable deliveryAuthorization, private runtime, evidence, reporting, and retest follow one engagement standard.
Systems engineering
Security properties should survive real operation.
Our systems work explores the places where trust is often handed away too casually: browsers, mail, privileged services, AI context, rebuilds, and recovery.
Selected systems work
✉
OSMAPSecure mail for high-consequence systems, with least privilege and content isolation.
Inspect →▦
OpenBSD MailstackReproducible hardening, operator control, and security that survives rebuild and recovery.
Inspect →◉
Browser-Safe AI SecurityLocal, reproducible investigation of prompt injection, source confusion, and browser-state abuse.
Inspect →Client-specific engineering
If the proof is missing, we build the instrument.
Validators, harnesses, bounded proofs of concept, evidence collectors, and retest paths are built when the security question requires them.
Purpose-built for the security question
?
InvestigateUnderstand the unusual condition.⌁
ResearchFind the missing mechanism.⚙
BuildCreate the minimum useful instrument.□
ValidateMake the condition observable.◇
DemonstrateShow engineers what must change.✓
RetestVerify the corrected condition.Bring us the difficult question
Bring the problem standard tooling cannot explain.
Test deeper. Build the missing validator. Engineer the control. Produce the proof.
Choose the response the problem needs
◎
Test deeperFollow the attack path.
⚙
BuildCreate the missing capability.
⌁
ResearchResolve the unknown mechanism.
✓
EngineerTurn the lesson into a durable control.