🤖 DAILY AI NEWS DIGEST

01@nvidia: Agents are becoming specialized, using reinforcement learning post-training to develop domain-specific skills ... 2026-07-17𝕏 X (official)

244 likes, 29 reposts | Agents are becoming specialized, using reinforcement learning post-training to develop domain-specific skills 🛠️ Every post-training rollout is an inference call, so a lower cost per token flows straight into a higher Intelligence per Dollar on every run. That is what the htt

x.com

02@OpenAI: GPT-5.6 Sol sets a new state of the art in cybersecurity on “The Last Ones” cyber range. We’re already seeing... 2026-07-17𝕏 X (official)

2236 likes, 183 reposts | GPT-5.6 Sol sets a new state of the art in cybersecurity on “The Last Ones” cyber range. We’re already seeing that capability translate into defensive outcomes: helping teams find, validate, and fix vulnerabilities in real-world code. Put it to work with Codex Security: htt

x.com

03@emollick: This is all interesting but specifically, I, too, am curious about this. The US & UK clearly see the model... 2026-07-17𝕏 X (trusted)

163 likes, 7 reposts | This is all interesting but specifically, I, too, am curious about this. The US & UK clearly see the models being released today from the closed labs as presenting genuine cyber risk (as well as offensive capabilities), it is interesting that China does not seem to believe

x.com

04Releases · openclaw/openclaw · GitHub 2026-07-17🔍 ExaAI

• ▸ OpenClaw 2026.7.2 introduces remote coding sessions enabling Control UI on cloud workers and terminal-based session resumption for Codex, Claude, OpenCode, and Pi.
• ▸ It adds native automation parity on mobile, including Android Voice Wake in foreground, and exposes camera, location, and notification features from headless Linux nodes.
• ▸ The release improves channel safety by preventing Telegram ingress loss after restarts, maintaining Signal control responsiveness, and restricting channel allowlists from granting owner access.
• ▸ Guided Control UI setup now supports configuring model providers, onboarding channels via a setup page, and selecting images and models during session creation.
• ▸ Packaging enhancements include Linux deb and AppImage bundles with Gateway guidance and streamlined Windows installs continuing immediately after winget adds Node.js.

github.com

05@simonw: "Early in the rollout of the technology, employees also faced restrictions on using Gemini to write or analyze... 2026-07-17𝕏 X (trusted)

359 likes, 12 reposts | "Early in the rollout of the technology, employees also faced restrictions on using Gemini to write or analyze software over concerns that proprietary code could leak into the AI model’s training data, they said." ... concerns about their OWN code being trained on? https://t.

x.com

06Netflix acquired Ben Affleck’s AI startup for nearly $600 million. | The Verge 2026-07-17🔍 ExaAI

• ▸ Netflix acquired Ben Affleck’s AI startup for approximately $587 million in March 2026.
• ▸ The acquired AI company specializes in generative AI technology, which Netflix has already used in around 300 titles.
• ▸ This acquisition aims to enhance Netflix’s content creation and streaming capabilities through advanced AI tools.

theverge.com

07Capital One releases VulnHunter, an open-source AI tool that finds software flaws before hackers do | VentureBeat 2026-07-17🔍 ExaAI

• ▸ Capital One launched VulnHunter, an open-source AI security tool that scans source code for exploitable vulnerabilities, maps attacker paths, and proposes targeted fixes before code reaches production.
• ▸ VulnHunter uses an "attacker-first forward analysis" approach, starting from likely attacker entry points and reasoning forward through application logic to identify real exploit paths, reducing false positives.
• ▸ The tool includes a built-in "falsification engine" that attempts to disprove its own vulnerability findings before presenting them to developers, providing detailed exploit explanations and proposed code fixes.
• ▸ VulnHunter currently runs on Anthropic's Claude Opus 4.8 model within a Claude Code environment but is designed to be compatible with other foundation models and coding frameworks.
• ▸ The tool is released under an Apache 2.0 license on GitHub to foster community collaboration and improve defenses against AI-driven cyber threats.

venturebeat.com

08BrainCo unveils a brain-to-robot platform at WAIC 2026-07-17🔍 ExaAI

• ▸ BrainCo launched the Brain-Controlled Robot AI Platform, an integrated brain-to-robot system that uses an EEG headset and AI to decode brain signals into robot commands within 200 milliseconds.
• ▸ The platform enables mind control of various off-the-shelf robots, including humanoids, robotic arms, and quadruped robots, without requiring special hardware.
• ▸ BrainCo introduced an Embodied AI Data Collection Solution that captures human demonstrations with a dual-arm rig, precision glove, and EEG to generate high-quality training data for embodied AI applications.

thenextweb.com

09DeepMind launches an AI biosecurity programme 2026-07-17🔍 ExaAI

• ▸ DeepMind and Isomorphic Labs launched a bioresilience AI programme to prevent, detect, and respond to biological threats using AI.
• ▸ The programme includes a four-step safety process (threat modelling, evaluations, mitigations, monitoring) and adapts SynthID watermarking for DNA synthesis to screen risky AI-generated sequences.
• ▸ AlphaEvolve agent accelerates pathogen sequencing algorithms to enable faster and cheaper outbreak tracking.
• ▸ Trusted researchers and biosecurity partners get restricted access to advanced models for vaccine design and drug discovery during outbreaks.

thenextweb.com

10Bunkerhill raises $55M for hospital-built clinical AI agents 2026-07-17🔍 ExaAI

• ▸ Bunkerhill Health launched Carebricks, a platform that enables health systems to build and deploy their own clinical AI agents instead of purchasing vendor-specific tools.
• ▸ Carebricks agents actively monitor unstructured data, prioritize cases, and route patients, moving AI from passive documentation to actionable clinical workflows.
• ▸ Over 20 AI agents are live at the University of Texas Medical Branch, including an FDA-cleared coronary calcium agent that flagged a high-risk patient, leading to life-saving intervention.
• ▸ Other agents reportedly reduced specialist wait times by over 50% and accelerated urgent lung nodule findings by 80%.
• ▸ Only some algorithms are FDA-cleared, and the platform raises important questions about clinical autonomy, liability, and oversight in AI-driven patient triage.

thenextweb.com

11Introducing PrismaScout -- PrismaQuant v2! - DGX Spark / GB10 User Forum / DGX Spark / GB10 - NVIDIA Developer Forums 2026-07-17🔍 ExaAI

• ▸ PrismaScout and PrismaQuant v2 are tools/scripts designed for automated testing and quantization of large language models, supporting recipes for models like qwen3.6-35b and ornith-1.0-35b.
• ▸ The tools enable generation of multiple test variants of recipe YAML files with configurable options such as prefix caching, speculative decoding, and bf16 precision.
• ▸ They include automation scripts (`run-test.sh`, `run-local-test.sh`, `generate-recipe.sh`) to run tests and generate quantization recipes, facilitating benchmarking and evaluation of model performance.
• ▸ The toolset supports integration with Docker environments and provides Python scripts for summarizing evaluation results.

forums.developer.nvidia.com