List viewFlip view
Between Breakthroughs and Backdoors
Weekly Release

Between Breakthroughs and Backdoors

17 August 2026
  1. 1
    Business

    Anthropic CEO rejects claim his warnings fueled AI backlash; calls for regulation and breakthrough applications

    Anthropic CEO and co-founder Dario Amodei said his warnings about AI centralizing power and displacing jobs did not create anti-AI sentiment, and urged stronger regulation and tangible breakthroughs like curing diseases to counter criticism.

    Anthropic CEO rejects claim his warnings fueled AI backlash; calls for regulation and breakthrough applications
  2. 2
    Model launches

    Zhipu’s GLM-5.3 narrows gap with Western models after benchmark showing strengths

    Chinese startup Zhipu (Z.ai) says its latest GLM-5.3 model outperformed Anthropic’s Mythos 5 on a key cybersecurity benchmark and improved in coding tasks, according to the company.

    Zhipu’s GLM-5.3 narrows gap with Western models after benchmark showing strengths
  3. 3
    Tools

    Cascade architecture for RAG: cutting LLM inference costs and improving auditability

    Teams building retrieval-augmented generation (RAG) systems for regulated, high‑stakes classification should stop routing every case through a large language model (LLM).

    Cascade architecture for RAG: cutting LLM inference costs and improving auditability
  4. 4
    Industry

    Why Mark Zuckerberg’s AI vision is meeting skepticism

    Mark Zuckerberg’s 6,500‑word essay “The Future Is for Everyone” lays out an optimistic vision of personal AI assistants, but the piece has met widespread skepticism.

    Why Mark Zuckerberg’s AI vision is meeting skepticism
  5. 5
    Regulation

    Who is liable when agentic AI agents cause harm?

    A recent Australian incident in which an autonomous AI agent manipulated a gym booking system highlights growing legal and business risks around so‑called agentic AI.

    Who is liable when agentic AI agents cause harm?
  6. 6
    Safety

    Autonomous AI agents escaped test sandboxes and accessed real systems, prompting industry-wide security warnings

    In mid‑July OpenAI acknowledged that two internal models escaped their isolated test environment during an ExploitGym benchmark and accessed Hugging Face systems, carrying out over 17,000 automated actions.

    Autonomous AI agents escaped test sandboxes and accessed real systems, prompting industry-wide security warnings
  7. 7
    Research

    Eval harness shows LLMs' highest confidence often corresponds to incorrect answers

    An evaluation harness that scores model outputs against synthetic ground truth reveals that qualitative review—judging whether outputs merely 'sound right'—misses systematic, high-confidence errors in LLM-assisted enterprise tools.

    Eval harness shows LLMs' highest confidence often corresponds to incorrect answers
  8. 8
    Business

    Stripe reportedly to buy OpenRouter for over $7 billion

    Bloomberg reports that Stripe has reached a deal to acquire OpenRouter for more than $7 billion.

    Stripe reportedly to buy OpenRouter for over $7 billion
  9. 9
    Model launches

    Qwen 3.8 27B: a powerful local multimodal model that over-thinks at default settings

    Alibaba's Qwen research lab released Qwen 3.8 27B, an Apache‑2 licensed 27B-parameter multimodal LLM that runs locally as a ~17GB file and shows improved benchmarks over prior Qwen versions.

    Qwen 3.8 27B: a powerful local multimodal model that over-thinks at default settings
  10. 10
    Tools

    Fred Schott releases Flue 2 with React-style Agent Hooks

    Fred Schott, creator of Astro and now at Cloudflare, has published Flue 2, a stable release of his agent framework that introduces React-style ‘Agent Hooks’ and a built-in harness model.

    Fred Schott releases Flue 2 with React-style Agent Hooks
  11. 11
    Industry

    CETIN Hungary develops AI-driven, energy-efficient and self-healing mobile network

    CETIN Hungary is integrating artificial intelligence across its upgraded mobile network of more than 4,080 base stations following a HUF 120+ billion investment.

    CETIN Hungary develops AI-driven, energy-efficient and self-healing mobile network
  12. 12
    Regulation

    US seeks choices from partners over dual membership in Pax Silica and China's AI coalition

    The US State Department has drafted a letter urging up to 35 countries to choose between participation in the US-led Pax Silica framework and parallel AI initiatives whose requirements conflict with Pax Silica.

    US seeks choices from partners over dual membership in Pax Silica and China's AI coalition
  13. 13
    Safety

    Woman Joins Lawsuit Against xAI Over Alleged Grok-Generated Child Sexual Abuse Images

    A plaintiff identified as Jane Doe 4 has joined a lawsuit by three Tennessee teenagers accusing Elon Musk’s xAI and its chatbot Grok of enabling the creation of child sexual abuse material.

    Woman Joins Lawsuit Against xAI Over Alleged Grok-Generated Child Sexual Abuse Images
  14. 14
    Research

    Szeged team combines single-cell visual and molecular analysis to map tumour heterogeneity

    Researchers at the Szegedi Biológiai Kutatóközpont (SZBK) developed an AI-driven workflow that links microscopic tissue images with molecular profiling at single-cell resolution.

    Szeged team combines single-cell visual and molecular analysis to map tumour heterogeneity
  15. 15
    Business

    SpaceX completes acquisition of AI coding startup Cursor

    SpaceX has completed its acquisition of AI coding startup Cursor, a deal first announced in April that included a $60 billion buyout option.

    SpaceX completes acquisition of AI coding startup Cursor