trust

Discusses the unresolved problem of AI agents being able to build working apps but remaining untrustworthy black boxes when deployed unattended in production.

An opinion piece exploring the trade-off between AI assistants that silently make assumptions and those that interrupt users for clarification, questioning the optimal level of initiative for tools like ChatGPT.

Lovable emphasizes that trust is built through invisible security measures, moving beyond 'vibe coding' to serious development.

A developer built an AI agent that drafts sales decks, but adoption was near zero until the agent showed its reasoning behind each slide choice. The lesson: making agent decisions visible is more critical for trust than raw output quality.

Raises security concerns about blindly trusting MCP server allowlists without reviewing tool schemas, comparing it to piping curl into bash in 2013.

In a Decoder podcast, Proton CTO Bart Butler discusses the company's philosophy of selling trust through privacy-centric products, the challenges of scaling while maintaining core values, and how it navigates government requests for user data, including a case where the Swiss government helped unmask a protester.

The article argues that in agent-to-agent communication, authentication alone is insufficient for authorization; instead, structured, inspectable claims about intent, identity, and authority are needed, with the human remaining the final authority.

An experiment tested AI agents on fake bookkeeping data with six planted errors; agents caught five, notably refusing to guess when data was insufficient, highlighting the value of uncertainty-awareness over pure accuracy.

The article explores methods for AI agents to authenticate their identity and prove whom they represent, addressing key trust and security challenges in autonomous systems.

Developers flagged a major privacy concern with Elon Musk's Grok AI model, leading to a trust crisis for the product.

The article discusses the concept of verifiable AI inference, exploring methods like trusted attestation and cryptographic proofs to ensure the authenticity and provenance of AI-generated outputs without rerunning the model.

NVIDIA's blog post highlights how open models like Nemotron enable enterprises and nations to build trustworthy, controllable, and customized AI systems, contrasting with closed models that limit inspection and adaptation.

A discussion on which boring internal tasks users would trust an AI agent to handle, and which tasks they would not.

A researcher demonstrates how to backdoor an open-weight coding model in under $100, raising concerns about trust in Chinese open-weight models like GLM 5.2.

An employee used an AI agent to auto-respond to Slack messages, and it gave a confidently wrong answer about a client deadline, highlighting the risk of trusting tone and fluency over accuracy.

The article argues that before AI agents can be widely deployed, they need verifiable identity and auditability to ensure trust and accountability. The ITU is working on international standards for this.

A pop quiz asking why AI labs pre-train their own models and design their own chips, with multiple choice answers ranging from unit economics to trust issues.

Discusses how the business model for AI agents may increasingly depend on establishing user trust after an initial click or interaction, highlighting the importance of post-click trust in AI-driven services.

This paper introduces psychological competence as a missing dimension in AI evaluation, proposing a conceptual framework to assess how AI systems support user cognition, emotional interpretation, and decision-making in human-facing roles.

This paper proposes an adversarial social epistemology framework for analyzing trust, deception, and inference chains in communicative landscapes involving humans and large language models, and outlines mechanisms for auditing trust breaches.