AI Safety & SecurityAI 安全與防護
202202 articles collected
This hub gathers the AI safety and security coverage on this site, spanning model safety (jailbreaks, alignment, system-card protections) and the extension of classic security into the AI era: prompt injection, supply-chain attacks, and account security.
The angle is practical: postmortems that break down real attack chains, tool-level write-ups of guardrails and defense ordering, and policy tracking focused on what new rules actually require of developers.
To build a baseline, start with the prompt injection and layered-protection picks below.
MUST READS
- 01Prompt Injection Is a Data-Trust Problem, Not a Prompt Problem
Hidden prompts in web pages turn scraped data into instructions, so builders must treat fetched content as untrusted input.
- 02How Anthropic Builds Multi-Layer Safeguards for Claude: A Blueprint for AI Product Teams
An inside look at Anthropic's Safeguards team: policy, training, testing, real-time detection, and monitoring—and what product builders can learn.
- 03OpenRouter Guardrails: Budget and Safety Gates for Agents at the Workspace Layer
OpenRouter Guardrails bundle budget enforcement, ZDR, model limits, prompt injection defense, and DLP into one no-code workspace layer, with per-entity budgets and inheritance that only tightens.
- 04Inside the TanStack npm Supply-Chain Compromise
A pwn request, a poisoned Actions cache, and an OIDC token dumped from runner memory let attackers publish 84 malicious versions of 42 TanStack npm packages on May 11, 2026.
- 05Claude's July incidents: What we changed in alignment and security
Anthropic details containment fixes, evaluator best practices, and alignment research after Claude models accessed real systems during cyber evaluations.
- 06AI-Enabled Cyber Threats: Why Old Security Frameworks Are Failing
Anthropic's year-long analysis of 832 banned accounts reveals how AI is making attackers more dangerous and why MITRE ATT&CK needs an update.
TOPIC TIMELINE
All 202 articles in this hub, newest first.
- Who Writes the Safety Rules for AI?AI Safety
- Prompt Injection Is a Data-Trust Problem, Not a Prompt ProblemPrompt Injection
- Fable 5.1 and Mythos 5.1: Same Model, Cache Reads 75% OffClaude Fable 5.1
- Gemini 3.8 Flash Intro Price Doubles on Jan 1: 54.9% HLEGemini 3.8 Flash
- Go 1.27 Lands: Generic Methods, json/v2, Post-Quantum TLSDeveloper Tools
- Docker Sandboxes: microVM Isolation for AI Coding AgentsDeveloper Tools
- Edge Ends Manifest V2: What uBlock Origin Users Do NowDeveloper Tools
- Domain-Camouflaged Injections Evade LLM Injection DetectorsPrompt Injection
- Google Scanned the Web for Indirect Prompt InjectionsPrompt Injection
- NVD Gives Up on CVE Backlog as AI Inflow AcceleratesCybersecurity
- WHO Experts Frame Generative AI as a Public Health IssueResponsible AI
- Microsoft Tests 60 Ways to Verify What's Real OnlineContent-Authenticity
- AI Found All 12 OpenSSL Zero-Days in One ReleaseCybersecurity
- CrowdStrike Buys SGNL to Secure AI Agent IdentitiesCybersecurity