Nobody's Steering: The Liability Question Nobody Wants to Answer

Analysis

Benchmarking Agent Reliability Under Adversarial Task Drift

New research digs into what happens when autonomous agents get fed slightly-off instructions mid-task — turns out most fold way faster than their demos suggest. If you're building a zero-human company, this is the boring paper you actually need to read before pitching investors on 'full autonomy.'

YC Partners Debate Where Founder Liability Actually Sits

A fresh round of YC commentary is circling the same uncomfortable question: if your company is run by agents, who signs the SAFE, who eats the lawsuit, who goes to jail if something breaks. Nobody's got a clean answer yet, which tells you the legal infrastructure is years behind the technical infrastructure.

Multi-Agent Coordination Failures at Scale

Researchers stress-tested agent swarms coordinating on shared goals and found failure rates climb fast past a handful of agents without a central arbiter. Basically confirms what anyone running a real agent stack already suspected — more agents doesn't mean more capability, it means more ways to fall over.

Revisiting Agent Individuality: Do Autonomous Orgs Need Personas?

Older paper getting fresh citations as builders debate whether giving agents distinct 'roles' actually improves org-level decision-making or just adds theater. Worth a skim if you're designing the org chart for your headless company — turns out personas might matter more for human trust than for agent performance.

Tools

AutoGPT Ships Tighter Guardrails for Financial Actions

Latest release locks down what agents can do with payment rails and API keys without an explicit human-signed policy file. Smart move — the second an autonomous agent moves real money badly, the whole 'zero-human' narrative takes a PR hit it won't recover from.

News

WalletConnect Pushes Deeper Into Agent-Controlled Treasuries

They're building out permissioning layers so agents can spend from a treasury without a human clicking 'approve' every time. Useful, but it's also exactly the kind of infrastructure that makes the liability question above way more urgent — someone's going to test these limits with real capital soon.