1sbang/mike

A focused security exploration of Mike, centred on keeping a legal assistant from revealing or misusing sensitive information.

Eight tracked security threads point to a focused exploration, but with no separate product shape yet and no recent momentum since the 5 May 2026 push.

View on GitHub →

1sbang's fork is exploring a tighter set of boundaries for Mike's legal-assistant conversations: protecting its confidential instructions, refusing attempts to extract personal information, and declining harmful requests involving tools or bulk data.

The work grew from automated adversarial testing that identified concrete weak spots. It is a security-focused direction rather than a new legal-practice product or rebrand, and curious readers can visit GitHub for the underlying validation work.

What's in it

Prompt confidentiality Aims to stop users from coaxing the assistant into revealing or reconstructing its private instructions.
Personal-data protection Sets clearer limits on requests intended to expose sensitive personal information.
Safer tool use Teaches the assistant to refuse harmful or inappropriate requests involving tools and bulk data access.
Adversarial testing Grounded in automated attack testing, with identified gaps used to shape the proposed safeguards.

Direction

security

Activity

Themed changes and pull requests touching this fork, newest first. Themed changes that haven't been turned into a public post yet still appear — they're real work even without a published writeup.

⛔ #38 Security hardening: system prompt confidentiality, PII boundaries, and tool use guardrails +35 6mo ago by 1sbang ↗ analysis ↗ GitHub

# Security hardening: system prompt confidentiality, PII boundaries, and tool use guardrails

⛔ #37 Security hardening: system prompt confidentiality, PII boundaries, and tool use guardrails +95,504 6mo ago by 1sbang ↗ analysis ↗ GitHub

# Security hardening: system prompt confidentiality, PII boundaries, and tool use guardrails

📝 1sbang hardens Mike against the questions it shouldn't answer 1 commit 6mo ago securitycompliance draft

A prompt-only rewrite that teaches the legal assistant to refuse leaking its own setup, handing over personal data, or being talked into bulk data grabs.

📝 1sbang adds adversarially-tested system prompt guardrails for PII and tool misuse 0 commits — securitycompliance draft

@1sbang ran 330 attack probes against Mike's baseline system prompt, found three concrete gaps, and submitted a fix. The PR closed unmerged 90 seconds after opening, but the work and validation artifacts are on the branc…

Threads of work (detailed view)

3 threads have been distilled into posts.

1sbang hardens Mike against the questions it shouldn't answer

A prompt-only rewrite that teaches the legal assistant to refuse leaking its own setup, handing over personal data, or being talked into bulk data grabs.

securitycompliance

1 commit · latest 6mo ago

System prompt hardening: PII refusals, tool misuse guardrails, prompt confidentiality

@1sbang ran 330 attack probes against Mike's baseline prompt and found three gaps worth fixing in a legal AI context. The resulting change is 35 lines added to `chatTools.ts` - no tool or routing changes, just prompt text.

securitychat-ui

1 commit · latest 6mo ago

1sbang adds adversarially-tested system prompt guardrails for PII and tool misuse

securitycompliance

Pull requests (detailed view)

2 PRs touch this fork — inbound (filed against it) or outbound (filed from it). State icons match the editorial dashboard.

⛔ Closed without merge (2)

⛔ closed · #38 Security hardening: system prompt confidentiality, PII boundaries, and tool use guardrails

by 1sbang · opened 6mo ago · closed 6mo ago

open-legal-products/mike ← 1sbang/mike · +35 · ↗ analysis · ↗ GitHub

⛔ closed · #37 Security hardening: system prompt confidentiality, PII boundaries, and tool use guardrails

by 1sbang · opened 6mo ago · closed 6mo ago

open-legal-products/mike ← 1sbang/mike · +95,504 · ↗ analysis · ↗ GitHub