Planet AI Weekly August 16, 2026

 

Planet AI Weekly

August 16, 2026

This week we reviewed 306 articles from 40 sources. 

NVIDIA shipped a full-duplex speech model with tool calling; Mistral and the White House both addressed AI sovereignty—one with compute, the other with policy. The week also brought a $5B Databricks round, a Grok abuse case, and OpenAI's internal reckoning over a rogue agent hack.

Official Highlights

NVIDIA's NemotronLabs VoiceChat 11B hits ~448 ms turn-taking latency as an open full-duplex speech-to-speech model with live tool calling—marking a concrete latency improvement over prior turn-based systems. Read more

Ollama v0.32.13 adds MLX model handling fixes and qwen3.8 developer instruction support, while LiteLLM v1.96.2 now cryptographically signs Docker images with cosign—both directly affect production deployments. Read more & Read even more

Mistral detailed European infrastructure commitments for sovereign AI, including in-region inference and open models, though specific compute capacity figures remain unspecified. Read more

The White House plans to expand its AI policy framework to include open models, according to sources familiar with the draft. Read more

Anthropic flipped Claude Code's auto mode on by default, reducing human review steps for routine programming tasks. Read more

From the Community

Dario Amodei argues the public's negative view of AI stems from a crisis of systemic trust, not warnings from AI leaders. Read more

A woman alleges her stepfather used Grok to transform a childhood photo into explicit imagery, spotlighting gaps in abuse prevention. Read more

Wired traces OpenAI's rogue agent hack and the internal safety reckoning it triggered, exposing cultural tension between shipping fast and locking down systems. Read more

Nathan Lambert argues Chinese labs aren't just distilling—they're keeping pace with frontier models in GLM-5.3. Read more

Embattled hedge fund Situational Awareness invested $400M in chip startup Source Foundry, signaling a shift in AI hardware supply chains as traditional venture models face pressure. Read more

Wired dismantles Mark Zuckerberg's 6,500-word AI manifesto as a case study in corporate vacuity and AI hype. Read more

MarkTechPost compares LLM observability platforms—Langfuse, LangSmith, Braintrust, Arize, and others—on tracing depth, evaluation, monitoring, and pricing. Read more

Unitree's $16,000 G1 robot has found viral fame as a 4-foot-tall performer, raising concrete questions about when affordable embodied AI graduates from novelty to workforce application. Read more

Featured This Week

Bookmark this if you've ever had to explain a RAG hallucination to a CFO. Miodrag Cekikj's walkthrough shows how to enforce retrieval-only responses in property insurance claims—architecture, FastAPI scaffolding, and the refusal-to-guess logic that prevents expensive wrong answers. Read more

Send comments and story tips to tips@planet-ai.net. If you found this useful, share it with a colleague.

This Week in AI

Additional stories worth scanning. Title only — click through for the full piece.

Some Claude users are mad that Anthropic’s new watermarks will catch them using it at their jobs, classes — TechCrunch · 2026-08-12

Accel closes oversubscribed $550M India fund within weeks, 19 months after its last — TechCrunch · 2026-08-11

As AI-led attacks multiply, OpenAI launches a new cyber model — TechCrunch · 2026-08-10

Anthropic shares more details about how Claude’s new watermarks will work — TechCrunch · 2026-08-15

Google will now allow users to remove visible watermark from its AI generations — TechCrunch · 2026-08-14

OpenAI launches ChatGPT desktop app for Linux — TechCrunch · 2026-08-11

Mark Zuckerberg’s AI manifesto is exactly why people don’t like AI — TechCrunch · 2026-08-10

SpaceX officially closes its Cursor acquisition — TechCrunch · 2026-08-15

Does Mark Zuckerberg really believe AI is ‘for everyone’? — TechCrunch · 2026-08-14

Writer introduces new AI model and upgraded harness to contain token costs — TechCrunch · 2026-08-13

Amazon will train on Twitch streamers’ content by default, unless they opt out — TechCrunch · 2026-08-12

Google’s Gemini app surges to one billion users — TechCrunch · 2026-08-11

Tech industry is buzzing after a Claude agent hacked into a gym — TechCrunch · 2026-08-10

Amazon Can Use Your Twitch Content to Train Its AI—Unless You Opt Out — WIRED · 2026-08-15

Tech Visionary Says the Big AI Labs Don’t Get What People Want — WIRED · 2026-08-14

Databricks wanted to raise $1B, investors wanted $15B. It settled on $5B at a $190B valuation. — TechCrunch · 2026-08-13

Brad Lightcap, OpenAI’s longtime COO, is leaving to ‘start something new’ — TechCrunch · 2026-08-11

Meta’s new Glimmer AI model offers a hint at Zuckerberg’s personal intelligence vision — TechCrunch · 2026-08-10

Kog is going deeper to squeeze more inference out of GPUs — TechCrunch · 2026-08-14

OpenAI introduces ‘Ultrafast,’ a new mode that makes GPT-5.6 Sol work at 14x the speed — TechCrunch · 2026-08-13

AI coding startup Cognition reportedly already in talks to raise at $40B valuation — TechCrunch · 2026-08-12

General Catalyst leads $1.1B round into 2-month-old River AI — TechCrunch · 2026-08-11

Discovered Materials is playing AI whack-a-mole to hunt cooler chips — TechCrunch · 2026-08-10

Hyperscalers might regret embracing natural gas if new forecast proves correct — TechCrunch · 2026-08-14

IBM partners with OpenAI to bolster enterprise AI push — TechCrunch · 2026-08-13

This newsletter supports planet-ai.net, a curated aggregator for AI tutorials and official updates. Curated by Keith Larson.