TLDR AI 2026-07-31

🕵️ PleaseFix: a new vulnerability class hijacking agentic browsers, chained with "Intent Collision" into full 0-click attacks (account takeover, data theft, remote code execution), live at Zenity's booth at Black Hat on 8/6. Just nominated for a Pwnie Award for "Best AI Security Research."Get session details →

🎭 AgentForger: one crafted link planted a fully autonomous, attacker-controlled agent inside a victim's org, before OpenAI patched it. See how it worked →

🧭 The framework: Zenity's CISO's Guide to Securing Agentic AI covers "least agency," runtime boundaries that hold even when model alignment and identity checks don't. Download the guide →

Thinking Machines has released Inkling-Small, a 276B-parameter mixture-of-experts model with 12B active parameters. The model retains Inkling's multimodal reasoning, variable thinking effort, and 1M-token context window while using substantially less compute.

OpenAI reduced GPT-5.6 Luna pricing by 80% and Terra pricing by 20%, while improving Sol's API speed. The changes extended efficiency gains across API usage, Codex, and ChatGPT Work subscriptions.

Gemini Robotics ER 2 enhances automation capabilities with advanced AI and LLM integration. This innovation improves robotic efficiency, making it pivotal for industries needing precise automation.

Cursor described how making development environments easier for agents to understand, run, and test helped cloud agents grow from authoring about 10% of merged pull requests to more than half.

Enterprise AI projects increasingly reach production when vendors prove value on live workloads, provide ongoing testing and iteration, and expose ROI. Successful deployments start with decomposable workflows that ship quickly and expand, rather than broad transformations with undefined success criteria.

Open-weight LLMs have now reached accuracy parity with closed models in regulatory and clinical tasks, at significantly lower costs. The ClinReg benchmark showed models like GLM 5.2 and Kimi K3 performing within one standard deviation of top proprietary models like GPT 5.6 Sol, at one-third of the cost. Different models displayed distinct error profiles, highlighting the importance of choosing models based on task-specific requirements rather than just ranking.

WASTE is an open source inference engine designed to run models whose weights are substantially larger than the memory available on the host machine. It is an initial step into a broader effort to make increasingly capable models available on more hardware. The project aims to provide people and organizations greater control over infrastructure costs, data privacy, availability, and deployment. The first fully supported model by WASTE is Kimi K3, which can run on a MacBook Pro with 64 GB of unified memory.

Users should be able to close an account, keep a session, and hand it to another model. Stateful storage should be optional, and hosted tools should be observable. Compaction should be readable, and agent communication should be auditable. Distillation should be a path by which capability becomes more available, not a reason for building ever higher walls.

MiniMax H3 is an open model that breaks the boundaries between tasks and modalities. It understands unified context across text, images, video, and audio. The model can generate up to 15 seconds of video at 2K resolution with native stereo sound. H3 excels at instruction following, accurate text and brand rendering, and V2V motion transfer. Early testing shows that it is ready for commercial content creation across a wide range of use cases.

The Gemini Live API enables low-latency, real-time voice and vision interactions with Gemini. It processes continuous streams of audio, images, and text to deliver immediate, human-like responses. The API creates a natural conversational experience for users. It can be used to build real-time agents for a variety of industries.

Agent Behavior is an open standard for defining and evaluating how an AI agent should behave across a whole trajectory. Each behavior spec is a Markdown file that describes the recurring conduct that makes the agent reliable. The spec gives reviewers, rubrics, scorers, and evals something concrete to measure against. It can be used to review traces, write eval cases, revise prompts or tools, and communicate intended agent conduct to teams.

China's Moonshot AI open-sourced Kimi K3 on July 27. Any government, company, or individual can now run the model on their own computers for free. They can also retrain the model however they want. The existence of a high-quality open model increases return on hardware investments because of a reduction or elimination of ongoing licensing costs.

AI faces a bottleneck with GPU utilization, echoing how airlines maximize aircraft use for profitability. Owning more GPUs doesn't ensure efficiency - companies must focus on optimizing GPU workload orchestration to maximize output. Specialized models and continuous GPU management are crucial strategies for improving utilization and staying competitive in the AI arena.

500,000 people have already signed up for TLDR Hardware, our new twice-weekly newsletter covering chips, robotics, energy, and devices. If you work in hardware and want to help curate it, send your LinkedIn or resume to hardware@tldr.tech!

Empower every team to build with AI while giving IT and security complete visibility, control, and governance.

Hugging Face Storage Buckets allow developers to store models, datasets, and artifacts with simple per-TB pricing.

Superlogical plans to build a composable multiplexer for all work that is safe and operable in production.

NVIDIA has identified configuration issues causing AI cluster performance gaps, such as CPU power settings and network tuning differences.

Anthropic discovered three instances where its models accessed the internet during an evaluation and gained unauthorized access to the systems of three different organizations.

A judge said the Trump administration had not provided enough evidence to classify Anthropic as a supply-chain risk or justify blocking federal agencies from using its technology.

Loka, Arcee, AWS, and Prime Intellect post-trained Trinity Mini with reinforcement learning across tool-assisted biomedical research and Gene Ontology annotation.