More Molly Guards
Summary
A follow-up post showcasing various examples of 'molly guards'—physical and software design elements that prevent accidental button presses or actions—ranging from IBM typewriters to Chrome's quit dialog.
View Cached Full Text
Cached at: 06/10/26, 05:43 AM
Similar Articles
OSGuard: A Benchmark for Safety in Computer-Use Agents
OSGuard is a dual-granularity benchmark for evaluating safety in computer-use agents under benign user instructions, featuring action-level judgments and risk-augmented execution suites to detect unsafe shortcuts.
@aacle_: Everyone's obsessing over prompt injection hiding in PDFs and websites. Meanwhile the tool list itself — the thing ever…
A tweet and blog post highlight that the tool list itself is an attack surface: by controlling a tool's description, an attacker can hijack an AI agent's behavior without exploit code. This is part 2 of a series on MCP bug bounties.
Glimmer seems pretty censored?
A user reports that Muse Glimmer refuses to write mouse-control code, citing safety concerns, even for legitimate debugging tasks.
A single invisible character disabled one of our guardrails for three weeks, and the symptom looked exactly like model flakiness
A developer recounts a three-week production bug where a regex with a literal backspace character silently disabled a language-detection guardrail, making the LLM appear flaky. The post highlights the need to instrument deterministic guardrails to distinguish them from model nondeterminism.
@gwenshap: One quirk of AI generated code is excessive guard rails. Recently, I wanted to test a new API with a local stack. I ask…
A developer shares an experience where OpenAI's Codex added an excessive guard rail by inserting a runtime extension existence check into an API, which a human engineer would never do.