Skip to content
Daily digest · August 12, 202612 stories · 6 sources

August 12, 2026

Today's top tech stories, deduped across the newsletters I read and briefly summarized. Click a source to open the original article.

AI

  1. Anthropic's Compliance API now returns transcripts of local Cowork and Claude Code sessions

    Anthropic's Compliance API now returns transcripts of Cowork and Claude Code sessions that run on users' own machines, in beta for Claude Enterprise organizations. New endpoints let organizations list and retrieve sessions for compliance and audit purposes.

  2. NVIDIA launches Nemotron 3.5 Lightning and NeMo Switchyard for AI agents

    NVIDIA introduced Nemotron 3.5 Lightning, a fast, accurate model built for specialized task execution in long-running agents, alongside NeMo Switchyard, a routing system that spreads agent workloads across multiple models based on strengths and cost.

  3. NVIDIA JetPack 7.2.1 adds agentic video skills and T3000 emulation

    The new JetPack 7.2.1 release for NVIDIA Jetson adds agentic video skills and T3000 emulation, strengthening the platform for robotics, industrial automation, and video analytics.

  4. Lovable: The model picker is a dead end

    Lovable argues that letting users manually pick between AI models is the wrong approach, and that products should instead automatically select the right model for each task.

  5. Meta releases Muse Glimmer, an open 30B model under Apache 2.0

    Meta is back in the open-weights game: Muse Glimmer is a brand-new 30 billion parameter agentic model released under a clean Apache 2.0 license, a step up from the company's more restrictive older Llama licenses.

  6. OpenAI's head of ethics leaves less than a year after joining

    OpenAI's head of ethics is departing less than a year into the role, raising fresh questions about how the company prioritizes responsible AI development internally.

Security

  1. Researchers steal reasoning traces from proprietary LLM APIs

    A new research paper shows how the hidden reasoning traces from Anthropic, OpenAI, and Google can be reconstructed from proprietary APIs, even though the providers withhold them from users.

  2. Microsoft's August updates fix 400 flaws, including three zero-days

    Microsoft's August 2026 Patch Tuesday fixes 400 vulnerabilities, including one actively exploited flaw and two publicly disclosed zero-days. Alongside it, Microsoft shipped cumulative updates for Windows 10 (KB5120249) and Windows 11 (KB5121003 and KB5120240).

  3. Sandworm hackers target IT pros with trojanized WireGuard VPN client

    The Russian threat group Sandworm has been luring system administrators and IT professionals with fake job offers since at least May, spreading a trojanized version of the WireGuard VPN client.

  4. Cisco warns of ASA and FTD VPN flaw exploited to crash devices

    Cisco is warning that a high-severity denial-of-service vulnerability in Secure Firewall ASA and Threat Defense software is being actively exploited to remotely crash affected devices.

Dev

  1. Mojo 1.0 has arrived

    Modular has released Mojo 1.0, the first stable version of the programming language designed to combine Python's ease of use with C++-level performance for AI and systems development.

  2. Google: Go is an ideal language for AI-assisted software engineering

    Google argues that Go's simple syntax, strong standard library, and predictable structure make it particularly well suited for AI-assisted code generation.