What changed
FACT: Hong Kong's regulator ordered crypto platforms and online brokers to deploy phishing-resistant login within 12 months (Cointelegraph source). FACT: computer-use browser automation is now available in a cheap Flash-tier model (DeepMind source) and OpenAI ships long-running multi-app agents (OpenAI source). INFERENCE: device-bound FIDO2 credentials cannot be typed or replayed by browser agents, so the mandate structurally breaks agent-driven access to these platforms.
Why now
The collision is dateable: a 12-month compliance clock started in the same quarter that computer-use agents became cheap. Platforms must ship passkey auth before agent-using customers churn or route around it.
Converging signals
Regulation (HK phishing-resistant login mandate, 12-month deadline) x AI (Gemini 3.5 Flash computer use; ChatGPT long-running agents). The bridge is real: the mandated auth mechanism is by design hostile to the newly-cheap access mode.
Customer pain
HYPOTHESIS: platforms fear locking out power users and B2B integrations that rely on browser automation; users fear losing agent workflows. NO demand_evidence was supplied — zero complaints, job postings, or spend signals confirm anyone currently experiences or budgets for this pain. The pain is projected, not observed.
Who pays
Nominally HK-licensed crypto platforms and online brokers. INFERENCE: these are regulated financial enterprises with security-vendor due-diligence processes (pen tests, SOC2/ISO, local presence, procurement committees) — the exact buyer class the founder profile excludes as sole channel.
Solved today
FACT-ADJACENT (well-known, not in sources): trading platforms already expose OAuth/API-key trading APIs, which is the sanctioned way for automation to access accounts without touching login at all. Agent-auth infrastructure is an active VC-funded category (Anon, Stytch Connected Apps, Auth0 'Auth for GenAI', Arcade.dev, Browserbase) — treat as hypothesis to verify, but multiple incumbents are building delegated agent credentials now.
Why current solutions are bad
API coverage is partial and browser agents work where APIs don't. But that gap is the platforms' own roadmap item: the natural fix is the platform issuing scoped agent tokens itself or buying from its existing auth vendor — not from a third-party broker sitting between the user's passkey and the account, which is a security anti-pattern a regulator just tried to eliminate.
Proposed product
A delegated-session broker SDK/service: human completes passkey ceremony once; broker mints narrow, time-boxed, revocable, fully-audited agent sessions (scoped to e.g. read-only portfolio or capped trade sizes); sold to platforms as compliance-plus-automation middleware.
MVP version
Demo against one sandbox brokerage: passkey login -> scoped token minting -> agent (computer-use or API) operating under the scoped session -> audit log + instant revocation console. Buildable in 60-90 days as a demo; production-grade security infrastructure for regulated finance is a different, much longer bar.
30-day build
Verify the premise before building: interview 10 HK platform engineers/compliance officers; check whether SFC guidance addresses delegated/agent access; map which platforms lack trading APIs.
60-day build
If (unlikely) validation succeeds: sandbox demo + security architecture whitepaper; begin SOC2 groundwork.
90-day revenue plan
Realistically none. Regulated financial platforms will not put an unaudited solo-founder credential broker in their login path within 90 days; first revenue would be a paid pilot at month 9-12+ after security review.
Distribution path
Weakest link: founder has no HK network, no financial-infrastructure credibility markers, and the buyer requires vendor due diligence. Demonstrated-value selling does not shortcut security procurement at regulated brokers.
Pricing hypothesis
HYPOTHESIS: platform-side licensing $2k-10k/mo per platform, or per-seat delegation fees. No evidence of willingness to pay was provided.
Technical difficulty
High. WebAuthn delegation is deliberately hard; the broker becomes a high-value honeypot holding live sessions to crypto trading accounts. A breach is existential for customer and founder.
Legal / regulatory risk
Severe. Sitting between a user's phishing-resistant credential and a regulated trading account may itself violate the spirit (or letter) of the SFC rule; liability for agent-executed trades and stolen sessions is unbounded. Cross-border (US founder, HK financial regulation) adds licensing/advice-of-counsel cost.
Platform dependency
Total: the product only works if each platform integrates it, and each platform can replicate the core feature (scoped agent tokens) internally or via its existing auth vendor.
Founder fit
Low. The government-portal lesson (confidence 0.80) does NOT apply: this mandate compels platforms to deploy auth, not a class of filers to submit to a portal — there is no per-filing transaction for a solo tool to capture. Buyer is regulated-finance enterprise security procurement, on the founder's avoid list. Capital-and-runway lesson (0.90) applied: the idea is not killed for upfront cost or ramp — it is killed for buyer reachability and defensibility.
Breakout potential
If agent-auth becomes standardized, the winners are the auth incumbents and the platforms themselves; a solo broker in the credential path has no durable wedge.
Final recommendation
KILL for this founder. The convergence analysis is genuinely insightful — the mandate-vs-agents collision is real and dateable — but the monetizable position belongs to platforms and incumbent auth vendors, the buyer is unreachable enterprise security procurement in a foreign regulated market, there is zero supplied demand evidence, and the liability profile (custodial sessions to crypto accounts) is disqualifying for a solo operator. This is not a forced-filer/portal-submission shape despite superficially involving a mandate.
Next action
Do not build. Log the underlying pattern (auth mandates breaking agent access) as a watch-item; redirect reasoning budget toward mandates that compel a class of filers to submit into a government system, per the high-confidence founder-fit lesson.