Wednesday, April 15
AI Accountability Lines Drawn
Anthropic is drawing sharp ideological lines against its closest rival. The company has publicly opposed an Illinois bill backed by OpenAI (opens in a new tab) that would exempt AI labs from liability even when their systems contribute to mass casualties or over $1 billion in property damage — a stance that lands just as Anthropic's rise is prompting some OpenAI investors (opens in a new tab) to quietly reconsider their positions. Meanwhile, an Anthropic co-founder confirmed the company briefed the Trump administration on Claude Mythos, the AI system that autonomously found zero-day security vulnerabilities, bringing frontier AI capabilities directly into the federal policy conversation. The liability fight and the Mythos briefing together signal that Anthropic is aggressively positioning itself as the safety-credible adult in the room while the regulatory window is still open.
AI is quietly dismantling hiring as a functional process. HR professionals in Metro Vancouver report that a flood of AI-generated job applications (opens in a new tab) has made traditional resume screening essentially useless, with so many candidates producing flawless-looking submissions that human reviewers can no longer distinguish signal from noise. At the same time, the ClawBench benchmark (opens in a new tab) — testing AI agents across 153 tasks on 144 live websites — found that even the best-performing model completed only 33.3% of everyday online tasks correctly, exposing a wide gap between AI's ability to game application forms and its ability to actually execute real-world workflows. The result is a lopsided disruption: AI is already breaking recruiting pipelines upstream, long before it can reliably replace the work being recruited for.
The digital infrastructure undergirding millions of small businesses took a quiet hit this week after someone planted backdoors in dozens of WordPress plugins (opens in a new tab) used across thousands of websites, a supply-chain attack that gives attackers persistent access with minimal forensic trace. Separate from that, Fluidstack — an AI data center startup — is reportedly in talks to raise $1 billion at an $18 billion valuation, having already been valued at $7.5 billion just months ago, a more-than-doubling that reflects how voracious demand for GPU capacity continues to outpace any rational pricing anchor. These two stories sit on opposite ends of the infrastructure stack: one showing how fragile the existing web remains, the other showing how much capital is racing to build what comes next.