Tag
OpenAI published a 34-page guide on building AI agents, emphasizing that an agent is essentially a loop: run the model, call a tool, feed back results, repeat until an exit condition. The guide covers tools, guardrails, and starting with a single loop before scaling to multiple agents.
The author proposes Automation Engineering as a discipline for designing triggers, guardrails, and success checks to make AI agents safe and reliable without constant human oversight.
This paper introduces LeanGuard, a lightweight bidirectional encoder-based safety guardrail that matches the accuracy of larger reasoning-based guardrails while being approximately 100x faster, challenging the assumption that chain-of-thought reasoning is necessary for effective moderation.
The author contrasts polished AI agent demos with the reality of production systems, noting that most agent code is for error handling and guardrails rather than the core intelligence.
A user expresses concern that current AI models have become less creative and more corporate-sounding due to safety guardrails, contrasting them with earlier open models that were more imaginative.
The article introduces a guardrail platform for AI agents that provides a control layer to block malicious prompts, hallucinations, risky actions, and cost spikes, enabling safe autonomous AI in business environments.
A discussion about real-world failures of autonomous AI agents in production, such as sending unauthorized emails, modifying records, deleting data, and spending money, seeking experiences and guardrails.
This article explores how Java development teams are establishing guardrails and best practices to manage the quality, security, and reliability of AI-generated code.
Box CEO Aaron Levie argues that AI agents will use software 100X more than people, requiring guardrails, authoritative data sources, logging, and collaboration features; platforms enabling headless interactions will be best positioned.
A deterministic guardrail tool is needed between an AI agent's decision and its irreversible actions such as trades or sending funds, to ensure safety.
The Trump administration demands unbreakable guardrails for Fable 5, a request described as impossible.
A team built an AI agent to automatically fix tech debt by scanning the codebase and opening PRs, finding that the hardest part was precisely defining the problem. They discuss challenges of running multiple agents on the same codebase and the need for guardrails.
Introduces Faramesh, an open-source runtime enforcement layer for AI agent tool calls that checks policies before actions run, offering a solution beyond observability or LLM-as-judge.
OrcaRouter is a new AI gateway that intelligently routes prompts to the best model, offering cost savings, guardrails, and full observability with zero token markup and a free tier.
A senior Google engineer released a free 421-page document covering agentic design patterns for AI systems, with code-backed chapters on prompt chaining, multi-agent coordination, guardrails, and reasoning.
The author released Bendex Arc, an open-source governance layer for AI agents that enforces authority, blocks manipulation, and includes a live demo for testing.
Anthropic's Claude Fable 5 safety guardrails were bypassed within 48 hours using techniques like Unicode substitution and multi-turn decomposition, highlighting weaknesses in stateless classifiers and the need for continuous adversarial testing.
Anthropic apologized for secretly throttling its new Claude Fable 5 model with hidden guardrails targeting distillation attempts, and will now make safeguards visible and route flagged queries to an older model instead.
Anthropic's new Claude Fable 5 model refuses to answer basic biology questions due to overly conservative safety filters aimed at preventing bioweapons misuse, highlighting the tradeoff between capability and safety.
The article discusses deploying AI agents in finance while ensuring compliance with NIS2/DORA regulations, focusing on transparency, guardrails, and accountability for potential data breaches.