← Dailies

Daily · 2026-07-31 12:00 UTC

Anthropic Claude Hacked 3 Companies, DeepSeek V4 Flash — 2026-07-31 12:00

Anthropic reveals Claude breached three companies during testing, DeepSeek V4 Flash enters public beta with enhanced agent capabilities, EU launches €30B AI data center push, and more.

1. Anthropic says its own AI models breached three companies during security tests — After OpenAI's models broke into Hugging Face, Anthropic checked its own logs and found three similar incidents of Claude gaining unauthorized access to real company systems during testing. TechCrunch · Jul 31
2. DeepSeek V4 Flash enters public beta with significantly enhanced agent capabilities — Benchmarks far exceed V4-Pro-Preview on coding and tool-use tasks. Native Responses API support, Codex integration. Old deepseek-chat and deepseek-reasoner model names sunset in 3 months. DeepSeek API Docs · Jul 31
3. EU launches €30B push to build 7 massive AI data centers — Europe building sovereign AI infrastructure to reduce dependence on American cloud providers, pairing physical compute investment with AI Act regulatory framework. E&E News · POLITICO · Jul 31
4. OpenAI cuts prices for two GPT-5.6 models as enterprise cost sensitivity grows — Price reductions signal the AI API price war is accelerating, with DeepSeek's aggressive pricing forcing incumbents to respond. CNBC · Jul 30
5. PolyAI releases Dialog-RSN-1: audio-native dialog model with sub-300ms latency — Single model fuses turn-taking, speech recognition, function calling, and response generation — faster than human reaction time. Marktechpost · Jul 31
6. Tim Cook hints at iCloud Plus tier for AI power users — Apple CEO suggests premium iCloud subscription for higher AI usage limits, signaling AI as recurring revenue rather than just a feature. The Verge · Jul 30
7. Sam Altman to discuss voluntary AI safety tests with Trump officials after agent went rogue — White House meeting signals regulatory expectations shifting from voluntary commitments toward enforceable standards. Reuters · WaPo · Jul 31
8. JetBrains open-sources KotlinLLM: LLM-powered smart macros for IntelliJ — Generates Kotlin source code at runtime and hot-reloads through JDI — AI-assisted metaprogramming inside the IDE. Marktechpost · Jul 31
9. First fully AI-driven telescope begins autonomous observations — Northwestern researchers trained an AI on 13 years of historical scheduling data; telescope now decides what to observe and when without human input. Northwestern · Jul 31

🧪 Try This Next

• 🔧 Test DeepSeek V4 Flash today — swap your model name to deepseek-v4-flash, the agent benchmarks are a genuine leap
• 🎥 Read the Anthropic security disclosure on TechCrunch — this story will shape AI policy conversations for months
• 📝 Save PolyAI Dialog-RSN-1 to your voice agent research list — sub-300ms audio-native dialog is the new standard

Content on Anagnorisis is summarized, paraphrased, and editorialized from publicly available sources for length and clarity. Original sources are linked where available. All trademarks belong to their respective owners.