Bloomsbury Tech · Hackathon Build · April 2026

Runtime risk control for autonomous payments.

Visa, Mastercard, and Stripe Radar were trained on a billion human transactions. AI agents settle in tenths of a cent through x402, delegate authority through tools, and read attacker-tampered invoices. We monitor, score, and govern financial actions taken by AI agents — before money moves.

100×

median tx compression Jun 2025 → Apr 2026

28,000×

daily volume growth, same window

99.6%

RF 5-fold accuracy on synthetic policy classes

97%

of features carry MI ≥ 0.05 (memo 08)

See it live →Methodology + memos

The risk vector — every action gets a six-dimension score before a decision.

Agent-likenessBehavioural driftPeer coordinationPolicy violationCounterparty riskPrompt-injection

Live Action Risk

Loading…

Behaviour map

PCA projection over the six sub-score risk vector (agent-likeness, drift, coordination, policy, counterparty, prompt-injection). Each point is one wallet; colour = tier; size = overall action risk.

Scenario:

Loading…

Tier counts: critical 0 · high 0 · medium 0 · low 0

Wallet leaderboard

Top 25 wallets by overall action risk — driven by the highest sub-score across agent-likeness / drift / coordination / policy / counterparty / prompt-injection.

Scenario:

Loading…