Key Highlights
- Safety turns biological. DeepMind and Isomorphic Labs unveiled a “bioresilience” strategy — extending SynthID watermarking to biological sequences, building cheaper pathogen surveillance, and standing up a rapid-response drug-design unit — framing frontier AI as both a biosecurity risk and the best defense against one.
- The open-model wave keeps cresting. Mira Murati’s Thinking Machines shipped its first open-weight model, Inkling (975B params, ~41B active, trained on 45T tokens), doubling down on the thesis that adaptable AI beats one-size-fits-all — while NVIDIA pushed Nemotron and Cosmos as “own your intelligence” infrastructure across a sweeping Japan rollout.
- AI is now writing the security news. Microsoft shipped a record 570 patches (two actively-exploited zero-days), explicitly crediting AI-assisted vulnerability discovery — the same week a Suno breach revealed the music generator had scraped YouTube, Deezer, and podcast feeds for training data.
- The money is moving to implementation and rivalry. Anthropic and Blackstone launched Ode, a $1.5B joint venture betting deployment services — not models — become the next trillion-dollar business, while Microsoft was reported training its salesforce to talk down Claude and OpenAI in favor of in-house Copilot models.
Analysis & Opinion
Our approach to bioresilience — Google DeepMind
DeepMind and Isomorphic Labs laid out a joint strategy for biosecurity in an era they say is being reshaped by ecosystem change, global connectivity, and the risk of AI misuse. The program rests on three pillars: prevention (threat modeling, external evaluations, and adapting SynthID watermarking to biological sequences), detection (cost-effective pathogen surveillance via algorithmic optimization and genome analysis), and response (giving vetted researchers advanced AI to speed vaccine and therapeutic development). Isomorphic has created a dedicated unit to deploy its drug-design capabilities rapidly during novel outbreaks, working with governments and international health bodies. The companies say they’ve built 15+ partnerships across government, biosecurity, and academia over the past year, tying the effort to their Frontier Safety Framework’s CBRN-risk mitigation. It’s a notably concrete safety proposal that treats dual-use directly — the same models that could aid bad actors are positioned as essential defensive tools.
Anthropic and Blackstone bet the next trillion-dollar AI business is implementation, not models — TechCrunch
Anthropic launched Ode, a $1.5B joint venture with Blackstone and other investors, to embed AI engineers directly inside customer organizations — acquiring Fractual AI and starting with 100 engineers. CEO Chris Taylor said it’s “pretty easy to imagine this as a trillion-dollar company someday if we execute well,” targeting enterprises where AI deployment is a CEO-level priority. Ode runs “Claude-first” but will use rival tools when needed; chief technologist Eddie Siegel argued “model selection matters, but it’s not where the majority of calories are spent.” The bet is that the durable value in enterprise AI is the messy work of integration, not the model itself — a thesis that puts Anthropic in direct competition with OpenAI’s The Deployment Company as well as Deloitte and Accenture.
Microsoft is reportedly training salespeople to talk down OpenAI and Anthropic — TechCrunch
At an internal meeting, Microsoft executives reportedly coached the salesforce to position competitors’ AI as slower, less accurate, and less secure than Microsoft’s own end-to-end stack — with a Copilot exec directly comparing Claude unfavorably on speed, accuracy, and Office security integration. EVP Jay Parikh framed it as “everyone else is selling parts — we’re selling the full end-to-end system.” It’s a striking reversal from Microsoft’s once-exclusive alliance with OpenAI, whose terms were loosened in April to let OpenAI serve other clients.
Hack suggests AI music generator Suno scraped YouTube for training data — TechCrunch
A supply-chain breach in November 2025 exposed Suno source code indicating the company scraped audio from YouTube Music, Deezer, Genius, stock libraries, and podcast RSS feeds for training. Suno has claimed fair use over “publicly available music,” but labels argue that circumventing YouTube’s anti-scraping protections violates the DMCA and terms of service. The same breach leaked customer emails, phone numbers, and partial card data stored in Stripe — and Suno reportedly never notified affected users. It sharpens two live fights at once: the copyright question over training data provenance, and the accountability gap when AI companies quietly sit on breaches.
Microsoft patches record number of security vulnerabilities, citing its use of AI — TechCrunch
Microsoft issued fixes for 570 flaws — its largest patch load ever — including two zero-days already being exploited (a Windows Server privilege-escalation bug and a SharePoint flaw CISA confirmed under active attack). The company had warned this was coming: “As AI helps defenders discover more issues, customers will see a higher volume of security updates.” Researchers are increasingly using AI to surface dormant bugs in decades-old code, making Windows’ vast legacy surface fertile ground. The double edge is obvious — the same capability arming defenders is available to attackers hunting the same latent flaws.
The DMA should not undercut security and privacy for Europeans — Google
Google argued that recent Digital Markets Act rulings “risk undermining vital privacy and security guardrails for millions of Europeans,” particularly around Android permissions and search-data sharing. The company warned that forcing third-party apps to access sensitive permissions without existing vetting could weaken device security, and that “Europeans’ private searches would be exposed to unfamiliar companies, without adequate anonymisation.” It proposed alternative compliance paths and called for evidence-based, flexible remedies targeted at documented harms. As with any incumbent’s regulatory pushback, the privacy framing cuts both ways — but it underscores how AI-assistant access to device data is becoming the new front in the DMA fight.
The three-second theft: why AI voice fraud outruns every defence — Hacker News
The piece argues detection-based defenses against AI voice cloning are already obsolete, opening with a Florida retiree who lost $15,000 to a call that perfectly mimicked her daughter’s voice pleading for bail money. Its core paradox: a scam requiring the “absolute frontier of machine learning” can be run against an ordinary grandmother, at scale, for essentially nothing. The FBI’s April 2026 report logged over 22,000 AI-related fraud complaints with losses exceeding $893M, and both law enforcement and banks have struggled for two years to contain it. The author’s conclusion is that the fix must shift from spotting fakes to hardening the human verification loop — because the cloning quality is no longer the weak link.
New Products & Tools
Thinking Machines launches its first open model, Inkling — TechCrunch
Mira Murati’s Thinking Machines Lab released Inkling, an open-weight mixture-of-experts model with 975B total parameters (~41B active per task), trained on 45T tokens across text, image, audio, and video — though outputs are currently limited to text, code, and structured data. It’s the lab’s first public showing after ~18 months mostly in stealth, and it’s positioned not as a standalone product but as a customizable foundation paired with its Tinker fine-tuning platform, doubling down on the thesis that adaptable models beat one-size-fits-all.
Apple Intelligence approved for launch in China with Alibaba’s Qwen — TechCrunch
China’s Cyberspace Administration cleared Apple Intelligence after Apple agreed to integrate Alibaba’s Qwen model into iOS, iPadOS, macOS, and visionOS, with Baidu also collaborating on features; no launch date was given, and Apple is reportedly exploring DeepSeek and ByteDance too.
Amid a hardware legal battle, OpenAI releases a $230 keyboard for Codex — TechCrunch
OpenAI and Work Louder launched the Codex Micro, a $230 light-up keyboard with agent-status keys, command shortcuts, a joystick, and a dial to tune how much compute agents spend — a limited-run “command center” for agentic coding, shipped as Apple’s trade-secret lawsuit against OpenAI looms.
NVIDIA introduces Jetson Thor for mainstream robotics and edge AI — NVIDIA
Two new Blackwell-based modules — the T3000 (865 FP4 TFLOPS at roughly half the size and power of the T5000) and the entry-level T2000 (400 TFLOPS) — target mainstream robotics, alongside agent skills that automated up to 15GB of memory savings for adopters like UBTech and Agile Robots.
Japan, industry leaders, and NVIDIA launch the world’s first national AI infrastructure — NVIDIA
Backed by Japan’s METI, NVIDIA and Noetra will build a 140MW AI factory with 13,750 Vera CPUs and 27,500 Rubin GPUs to power the national “FRONTia Project,” developing foundational models for agents, digital twins, and physical AI that will be made broadly accessible to domestic developers.
Japan’s robotics and manufacturing leaders build on NVIDIA Cosmos — NVIDIA
NVIDIA introduced Cosmos 3 Edge for on-device vision reasoning and robot-policy deployment on Jetson, and said FANUC, Sony, Yaskawa, and others plan to join the Cosmos Coalition to build open world models for testing AI before real-world deployment.
Japan’s enterprises and startups build industry-specialized AI with Nemotron — NVIDIA
Japanese groups are fine-tuning NVIDIA’s open Nemotron models for local needs — Institute of Science Tokyo’s Swallow, SB Intuitions’ Sarashina (selected by Japan’s Digital Agency), plus Hitachi and Sakana AI — under Jensen Huang’s pitch that “every nation and every company should own and control its intelligence infrastructure.”
Applied Computing wants to give oil and gas operators an AI model for the entire plant — TechCrunch
The London startup raised a $20M Series A (led by KBR, with Databricks Ventures) for Orbital, a foundation model blending time-series, physics-based modeling, and language processing to predict plant states — targeting facilities that today act on less than 8% of their available data.
Rime picks up $24M Series A to help enterprises field customer calls — TechCrunch
Voice-AI startup Rime raised $24M (M13 Ventures, Twilio Ventures) to build speech-to-speech models trained on proprietary studio-recorded data, using a phoneme-based architecture tuned for brand-name and industry-term pronunciation.
Whatnot acquires Shaped to power real-time live shopping recommendations — TechCrunch
The livestream-shopping platform bought ML recommendation startup Shaped to push personalization closer to real time — a hard problem when inventory, auctions, and buyer intent shift by the second.
Reelful’s AI turns your camera roll into short-form videos — TechCrunch
The iOS app, from ex-Snap ML engineer Kate Deyneka, takes a text prompt and a 30-second voice sample, then plans, scripts, voices, and assembles TikTok/Reels-ready videos from your photos and clips — with watermarks on all AI-generated output.
Build a multi-camera 3D tracking application with NVIDIA DeepStream 9.1 — NVIDIA Developer
DeepStream 9.1 adds AutoMagicCalib (automated camera calibration from existing footage) and Multi-View 3D Tracking to fuse detections across cameras into one 3D coordinate space, exposed through 13 agentic skills and JetPack 7.2 support.
Research
Develop lightweight USD runtimes faster with AI agents — NVIDIA Developer
NVIDIA’s nanousd-labs approach has AI agents read the OpenUSD Core Specification directly and generate spec-compliant runtime code validated against a spec-derived test suite — letting agents handle mechanical parsing and scene composition while engineers own architecture and performance tradeoffs.
Building faster cryptography with carryless multiplication in NVIDIA CUDA 13.3 — NVIDIA Developer
CUDA 13.3’s new hardware clmad instruction (Ampere and newer) accelerates carryless multiplication, hitting ~6.3 TB/s GHASH on a B200 (up to 18.8x faster than bitsliced implementations) and delivering 3–13x speedups for zero-knowledge proofs and sum-check protocols in binary extension fields.
References
- Our approach to bioresilience — Google DeepMind, 2026-07-16 [blog]
- Anthropic, Blackstone bet the next trillion-dollar AI business is implementation, not models — TechCrunch, 2026-07-15 [blog]
- Microsoft is reportedly training salespeople to talk down OpenAI and Anthropic — TechCrunch, 2026-07-15 [blog]
- Hack suggests AI music generator Suno scraped YouTube for training data — TechCrunch, 2026-07-15 [blog]
- Microsoft patches record number of security vulnerabilities, citing its use of AI — TechCrunch, 2026-07-15 [blog]
- The DMA should not undercut security and privacy for Europeans — Google, 2026-07-16 [blog]
- The three-second theft: why AI voice fraud outruns every defence — Hacker News, 2026-07-15 [blog]
- Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling — TechCrunch, 2026-07-15 [blog]
- Apple Intelligence approved for launch in China with Alibaba’s Qwen AI — TechCrunch, 2026-07-16 [blog]
- Amid hardware legal battle, OpenAI releases a $230 keyboard for Codex — TechCrunch, 2026-07-15 [blog]
- NVIDIA Introduces New Jetson Thor Computers to Advance Mainstream Robotics and Edge AI — NVIDIA, 2026-07-15 [blog]
- Japan Government, Industrial Leaders and NVIDIA Launch the World’s First National AI Infrastructure — NVIDIA, 2026-07-16 [blog]
- Japan’s Robotics and Manufacturing Leaders Build on NVIDIA Cosmos to Advance Physical AI Frontier — NVIDIA, 2026-07-15 [blog]
- Japan’s Enterprises and Startups Build Industry-Specialized AI With NVIDIA Nemotron Open Models — NVIDIA, 2026-07-15 [blog]
- Applied Computing wants to give oil and gas operators an AI model for the entire plant — TechCrunch, 2026-07-15 [blog]
- Rime picks up $24M Series A to help enterprises field customer calls — TechCrunch, 2026-07-15 [blog]
- Whatnot acquires Shaped to power real-time live shopping recommendations — TechCrunch, 2026-07-15 [blog]
- Reelful’s AI turns your camera roll into short-form videos for social media — TechCrunch, 2026-07-15 [blog]
- Build a Multi-Camera 3D Tracking Application with NVIDIA DeepStream 9.1 Skills — NVIDIA Developer, 2026-07-15 [blog]
- Develop Lightweight USD Runtimes Faster with AI Agents — NVIDIA Developer, 2026-07-15 [blog]
- Building Faster Cryptography with Carryless Multiplication in NVIDIA CUDA 13.3 — NVIDIA Developer, 2026-07-15 [blog]