Conduct

Guardrails for LLM and MCP tool calls

devtoolsintermediateFree tier

Overview

Conduct is an open-source guardrails layer for **LLM** and **MCP** tool calls that appeared on Show HN in late August 2026. It lets developers define policies, constraints, and permission schemas for how models may invoke tools, making it easier to enforce safety, compliance, and resource usage rules. The stack is likely centered on TypeScript/Node with JSON schemas, and it matters because it gives AI app builders a reusable policy engine instead of hand-rolled checks inside every agent or orchestration framework.

Best For

+ml-engineers
+backend-developers
+platform-engineers

News & Updates

METR Announces New Funding to Expand AI Evaluation Research

METR, a nonprofit focused on evaluating advanced AI systems, announced a new funding round to expand its research on measuring AI capabilities and risks. The organization, known for developing rigorous benchmarks for AI models, will use the funds to grow its team and develop more comprehensive evaluation methods. This investment comes as governments and companies increasingly rely on third-party assessments to understand AI safety and performance. The funding will enable METR to conduct more extensive testing of frontier models, potentially influencing regulatory standards.

deals8/30/2026

Anthropic's Compute Costs: Will Financing Bottleneck AI Scale?

A new case study from Epoch AI examines whether financing, not just chip supply, will bottleneck AI compute, using Anthropic as a key example. The analysis highlights that training frontier models like Claude requires billions in capital for GPUs and data centers, with Anthropic reportedly spending over $1 billion per training run. Epoch AI projects that by 2027, the compute needed for top models could exceed $100 billion, straining even major investors like Google and Amazon. This raises the question of whether capital markets, rather than semiconductor fabs, will become the limiting factor for AI progress.

breakthroughs8/29/2026

Conduct: Open-Source Guardrails for LLM and MCP Tool Calls

Conduct is a new open-source library that adds guardrails to LLM and MCP (Model Context Protocol) tool calls, ensuring safer and more controlled AI interactions. It allows developers to define validation rules, rate limits, and permission checks before a model executes a tool, preventing harmful or unintended actions. The project is hosted on GitHub and is designed to integrate seamlessly with existing LLM workflows, offering a lightweight alternative to heavier governance frameworks. This addresses a critical gap in AI safety, as tool-calling models become more common in production.

new_tools8/28/2026

Google DeepMind Tests Tamper-Proof AI Benchmarks with Singapore Safety Institute

Google DeepMind is piloting a double-blind evaluation of a frontier AI model for the first time, using cryptographic protection via Confidential Space to prevent Google from seeing test questions and evaluators from seeing model weights. The pilot, conducted with the Singapore AI Safety Institute, uses a Gemini Flash Lite model and aims to establish a new standard for tamper-proof AI benchmarks. This addresses growing concerns that AI benchmarks are unreliable due to data contamination and gaming. If successful, this approach could become the industry norm for independent AI safety evaluations.

breakthroughs8/28/2026

Russian Hackers Weaponize SpaceX's Cursor AI Agent in Cyber Breaches

Russian-speaking hackers have exploited SpaceX's Cursor AI agent to bypass safety guardrails and execute multiple cyber break-ins. The attackers leveraged the AI's code-generation capabilities to automate malicious scripts and evade detection. This marks a new frontier where AI development tools become dual-use weapons, enabling even low-skilled criminals to conduct sophisticated attacks. The breaches underscore the urgent need for AI firms to harden their models against adversarial misuse.

ai_society8/27/2026

Nvidia Launches PAC to Boost DC Influence as AI Chip Maker Expands Lobbying

Nvidia has launched a political action committee (PAC) as the AI chip giant ramps up its influence in Washington, D.C. The company, led by CEO Jensen Huang, is navigating increasing regulatory scrutiny over its dominant position in AI accelerators, with its H100 GPUs powering most large-scale AI training. The PAC will support federal candidates aligned with Nvidia's interests, including those on tech policy, export controls, and semiconductor manufacturing. This move comes as Nvidia faces potential export restrictions on advanced chips to China and growing antitrust attention from regulators. The PAC signals Nvidia's shift from a pure hardware vendor to a major policy player, aiming to shape legislation affecting AI and chip supply chains.

hardware8/27/2026

Nvidia Forms PAC to Boost Chips and AI Policy Influence in Washington

Nvidia has established a political action committee (PAC) to strengthen its lobbying efforts on chip and AI policy in Washington, D.C. The PAC will enable Nvidia to pool employee contributions and donate to political candidates who align with its interests. This move comes as Nvidia faces increasing scrutiny over its dominance in AI chips and export controls on advanced semiconductors to China. By forming a PAC, Nvidia aims to shape legislation on AI regulation, export policies, and domestic chip manufacturing incentives.

hardware8/27/2026

Pharma Companies Race to Build Drug Labs in Space as Zerodha Posts Flat FY26

The article reports on the growing race to manufacture pharmaceuticals in space, leveraging microgravity to grow protein crystals and other compounds that are difficult to produce on Earth. Companies like Bristol Myers Squibb and Merck have conducted experiments on the ISS, and startups such as Varda Space Industries are building orbital factories to return drug products to Earth. Meanwhile, Indian brokerage Zerodha reported a flat FY26, with revenue and profit largely unchanged from the previous year, highlighting the maturity of its business. The space drug race is driven by the potential for higher-quality crystals and novel formulations, but high costs and regulatory hurdles remain.

breakthroughs8/27/2026

Progressive Policy Institute Names Astri Kimball Van Dyke Senior Fellow for AI Policy

The Progressive Policy Institute (PPI) has appointed Astri Kimball Van Dyke as its Senior Fellow for Artificial Intelligence Policy. Van Dyke, who previously served in the Biden administration as Deputy Assistant to the President and Coordinator for the CHIPS Act, will lead PPI's work on AI governance and innovation. Her role will focus on shaping policy recommendations that balance technological advancement with public interest, drawing on her experience in semiconductor manufacturing and economic policy. This move signals PPI's commitment to influencing AI regulation as Congress debates new legislation.

ai_society8/27/2026

Researchers Develop Framework to Evaluate 10 Therapeutic Techniques Used by AI Chatbots in Counseling

Researchers Afonso Baldo, Hugo Pitorro, Areti Vassilopoulos and colleagues introduced an ontology of ten therapeutic moves to measure and steer how LLMs conduct psychotherapy. The moves are compact, function-based categories grounded in the MULTI-60 inventory, validated through analysis of AI counseling interactions. This framework enables systematic evaluation of AI's therapeutic techniques, addressing a critical gap as users increasingly rely on LLMs for emotional support. The work provides a foundation for improving AI's counseling quality and safety.

breakthroughs8/24/2026

Loading AI Digest

Overview

Best For

News & Updates

METR Announces New Funding to Expand AI Evaluation Research

Anthropic's Compute Costs: Will Financing Bottleneck AI Scale?

Conduct: Open-Source Guardrails for LLM and MCP Tool Calls

Google DeepMind Tests Tamper-Proof AI Benchmarks with Singapore Safety Institute

Russian Hackers Weaponize SpaceX's Cursor AI Agent in Cyber Breaches

Nvidia Launches PAC to Boost DC Influence as AI Chip Maker Expands Lobbying

Nvidia Forms PAC to Boost Chips and AI Policy Influence in Washington

Pharma Companies Race to Build Drug Labs in Space as Zerodha Posts Flat FY26

Progressive Policy Institute Names Astri Kimball Van Dyke Senior Fellow for AI Policy

Researchers Develop Framework to Evaluate 10 Therapeutic Techniques Used by AI Chatbots in Counseling