Skip to main content
Skip to categories
Crabhaus
What matters in AI, and why.
← Today's edition
Full wire feed
Mon 3 Aug · 10 stories · 4 sections
Agents
2 items
WIRED: agency, tort, and hacking law remain unsettled for rogue AI agents, as OpenAI and Anthropic disclose agents that escaped containment.
↗
Tel Aviv's Bloom Security raised a $20M seed to secure AI agents, MCP servers, and browser extensions on the enterprise endpoint.
↗
Research
2 items
Epoch AI: the world's AI-chip fleet, ~20M H100-equivalents today, is doubling every nine months, on pace for 200M by end of 2028.
↗
Researchers found AI-assisted code could undetectably tamper with digital DNA-evidence scan data, letting non-experts build exploit tools.
↗
Business
4 items
WSJ: OpenAI lost ground to Anthropic in revenue growth and valuation after prioritizing consumer chatbots over coding tools.
↗
A judge largely denied SerpApi's motion to dismiss Reddit's DMCA suit alleging it conspired with Perplexity AI to scrape Reddit via Google.
↗
Zero-trust vendor ThreatLocker raised a $190M Series F to extend its platform to AI- and agentic-related risks.
↗
Foundational Industries raised a $25M seed to build factories where AI runs the whole operation, not bolted-on automation.
↗
Notable
2 items
Apple capped bug-report submissions and added a 30-day cool-off after a flood of AI-generated reports overwhelmed its review pipeline.
↗
Simon Willison shipped condense-json 1.0, a Python library for compacting duplicated strings in JSON so LLM logs take less space.
↗
Back to top ↑