Episode Details

Please support this podcast by checking out our sponsors:
- SurveyMonkey, Using AI to surface insights faster and reduce manual analysis time - https://get.surveymonkey.com/tad
- Effortless AI design for presentations, websites, and more with Gamma - https://try.gamma.app/tad
- KrispCall: Agentic Cloud Telephony - https://try.krispcall.com/tad


Support The Automated Daily directly:
Buy me a coffee: https://buymeacoffee.com/theautomateddaily

Today's topics:

Claude Code auto-mode exploit - A reported prompt-injection attack on Claude Code auto mode raises fresh AI security concerns. The key issue is not just compromise, but that safety controls may have interfered with containment and cleanup.

Science benchmark tests AI agents - Terminal-Bench-Science 0.1 measures AI agents on real scientific workflows, not textbook questions. Early results show top models like Claude Opus 5 still solve only a minority of expert-curated research tasks.

DeepMind pilots blind model testing - Google DeepMind says it piloted a double-blind evaluation for a frontier AI model using cryptographic protections. The goal is to reduce benchmark contamination while preserving both model secrecy and test confidentiality.

OpenAI and Anthropic revenue surge - Epoch AI says OpenAI and Anthropic revenue growth may be the clearest signal of frontier AI adoption. Combined annualized revenue near $105 billion suggests generative AI is now a major economic force, not just a research trend.

Nvidia and DeepSeek escalate race - Nvidia's latest outlook points to extraordinary AI infrastructure demand, while DeepSeek reportedly nears a massive funding round. Together, the stories highlight how capital, compute, and a few dominant players are shaping the next phase of AI competition.

AI startups spread geographically - Stripe data suggests new AI-era businesses are forming farther from dense city centers, with more activity in smaller metros and outer suburbs. Even so, frontier AI research and top labs remain heavily concentrated in places like San Francisco.

Gemini flags counterfeit goods - An experiment using Google Gemini to spot fake cosmetic packaging showed both promise and unreliability. AI could catch subtle counterfeit clues, but false positives and authentic packaging errors make human verification essential.

Book scanning satire stings - A satirical essay about destroying antique books after scanning them for AI hits a real nerve around copyright, preservation, and tech culture. It matters because it mocks the way efficiency language can sanitize cultural loss.

-Claude Code Auto Mode Vulnerable to Prompt Injection
-Terminal-Bench-Science Launches Continuous Benchmark for Scientific AI Agents
-Google DeepMind Pilots Double-Blind AI Model Evaluations
-Epoch AI: OpenAI and Anthropic Revenue Growt