The Brief Sign in S

Latest

The newest stories across every desk. Read freely — sign in only if you want to keep bookmarks.

Thursday, 17 September 2026
No. 169

QuantHealth raises $45M to simulate trials before patients enroll

Business Wire

Tel Aviv clinical AI firm QuantHealth closed a $45 million Series B led by Qumra Capital, with Sanofi Ventures, Pitango HealthTech, and Bertelsmann Healthcare Investments among the backers, per its Business Wire release. The platform simulates trial designs on large real-world datasets so sponsors can stress-test protocols before enrollment. QuantHealth says it already works with 12 of the top 20 global drugmakers. Why it matters: many late-stage failures start as design mistakes that better simulation could catch earlier. Caveat: a funding round is not peer-reviewed proof that simulated wins become real Phase III wins.

No. 170

MiniMax opens H3 weights for omni-modal video, with 2K still gated

MiniMax

Shanghai's MiniMax open-sourced MiniMax H3 Base, an omni-modal video model that jointly reads text, images, video, and audio and can generate about 4 to 15 second clips with native stereo sound. First/last-frame and reference-conditioned checkpoints are on Hugging Face under a MiniMax community license, while the Context-IR layer and 2K regenerate module stay closed and API-only. Why it matters: an open Chinese video stack is now competing near the top of public generation and editing leaderboards. Caveat: full 2K still needs the closed module, and the license restricts commercial use for larger firms.

No. 171

InRisk Labs banks $27M as EarthRe wins a GIFT City reinsurance licence

The Economic Times

GIFT City climate insurtech InRisk Labs raised $27 million in a Series A co-led by Bessemer Venture Partners and Northpoint Capital, valuing the group near $70 million, cofounder Siddesh Ramasubramanian told The Economic Times. Most proceeds will capitalise subsidiary EarthRe Insurance IFSC, which won a final IFSCA reinsurance licence as GIFT City's first locally incorporated reinsurer. The stack mixes catastrophe models and AI underwriting with parametric payouts tied to weather indices. Why it matters: India needs regulated reinsurance capital for climate risk, not another demo. Caveat: AI here is still mostly modelling tooling, and scale in a crowded reinsurance market is unproven.

No. 172

Zuckerberg apologises to India over deepfakes, CSAM, and platform errors

The Hindu

After Meta's Joel Kaplan met MeitY officials and IT Minister Ashwini Vaishnaw on August 5, government sources said Mark Zuckerberg conveyed apologies for child sexual abuse material, deepfake content, and platform operating errors. Kaplan separately apologised for Facebook briefly restricting Prime Minister Narendra Modi's July 23 video. Officials argued Meta's algorithmic curation means Section 79 safe harbour should not apply, and said Meta admitted paid promotion of illegal content. Why it matters: India is linking synthetic media failures to intermediary liability. Caveat: Meta has not confirmed Zuckerberg's wording publicly, and lawyers say safe harbour is usually decided in court, case by case.

No. 173

Autonomous oxygen AI keeps hospital patients in target range longer

University of Colorado Anschutz

A University of Colorado Anschutz-led randomized trial of 300 adults found the O2matic PRO100 autonomous oxygen system kept patients in their target range 85 percent of the time, versus 63 percent with usual clinician titration. Published in JAMA Internal Medicine, SAVE-O2 AI also reported less hypoxemia and hyperoxemia, fewer manual adjustments, and no rise in serious adverse events across four U.S. hospitals. Why it matters: oxygen is ubiquitous, yet still mostly managed by intermittent manual tweaks. Caveat: the device was investigational in the United States for this study, and claimed workload benefits still need dedicated follow-up trials.

No. 174

DeepX edge chips jump to a $2.2B mark on a $30M Series D slice

Bloomberg

South Korean AI chip designer DeepX closed the first slice of a Series D that values the Pangyo startup at about $2.2 billion, roughly four times its prior mark, Bloomberg reported. Existing backers BNW Investment and DS Asset Management put in about $30 million. DeepX builds low-power AI semiconductors for robots, cameras, and other devices outside data centers, and said in April it had more than 30 mass-production contracts across eight countries. Why it matters: investor appetite is spreading from GPU racks into edge inference silicon. Caveat: the disclosed tranche came only from existing investors, and a larger close by September is still unfinished.

No. 175

ByteDance Seedance 2.5 makes 30-second audio-video the default clip length

ByteDance Seed

ByteDance's Seed team launched Seedance 2.5, a video model that generates up to 30 seconds of synchronized audio and video in one pass and can extend clips across rounds for longer stories. Creators may supply up to 30 images, 10 video clips, and 10 audio references, with tighter timestamp editing and camera control for ads and short films. It is rolling out on Jimeng AI and Doubao Pro, with API access promised on BytePlus ModelArk. Why it matters: China is pushing generative video toward controllable one-take storytelling. Caveat: complex multi-subject physics remains uneven, and copyright fights over training data are unresolved.

No. 176

CDSCO pilots AI while building one digital spine for drug regulation

The Hindu BusinessLine

India's drug regulator is piloting artificial intelligence across CDSCO workflows while preparing the first phase of an end-to-end digital platform within about 18 months. Drugs Controller General Rajeev Singh Raghuvanshi said the system would link research, manufacturing, approvals, distribution, and post-market surveillance. CDSCO already claims more than 99 percent of processes are digitised and plans QR checks on GMP and COPP certificates. Why it matters: New Delhi is treating AI as infrastructure for how medicines are cleared. Caveat: the AI pilots are still lightly specified, and the platform remains a timeline, not a finished rollout.

No. 177

Gracenote puts India at the center of its LLM entertainment metadata stack

The Economic Times

Nielsen unit Gracenote said its India technology centre now leads the company's AI product work, including a Model Context Protocol server that feeds large language models authoritative entertainment metadata for streaming and connected-TV discovery. Proprietary AI workflows, it says, turn unstructured title data into structured catalogues in minutes rather than days across more than 50 million titles. Why it matters: a major metadata vendor is grounding consumer AI search in curated catalogues built from an India hub. Caveat: this is still a company positioning note, and public proof of LLM accuracy gains from the MCP server remains thin.

No. 178

FDA clears Caristo's AI to score heart-attack risk from coronary inflammation

MedTech Dive

Caristo Diagnostics, an Oxford spinout, won FDA de novo authorization for CaRi-Heart, an AI tool that reads routine coronary CT angiography to quantify vessel inflammation and estimate 10-year cardiovascular mortality risk. The aim is patients with chest pain who lack obstructive coronary disease, a group linked to many major cardiac events in U.K. data on more than 40,000 CCTA patients. Caristo plans a U.S. commercial launch this quarter, using reimbursement codes that began in January. Why it matters: U.S. regulators green-lit inflammation imaging AI beyond plaque and stenosis scores. Caveat: outcome evidence is largely observational U.K. cohorts, not a finished U.S. outcomes trial.

No. 179

Tianjin lab's REAP robot loop lifts enzyme activity by tens of fold

Nature Communications

In Nature Communications, Tianjin University researchers introduce REAP, a closed-loop platform that pairs AI with robotic experiments to redesign enzymes. RankReg, a hybrid ranking-and-regression loss, learns from sparse wet-lab feedback and steers searches from single mutations into combinatorial variants, including distal sites beyond classic catalytic pockets. On cytochrome P450 BM3 the system reported a 57-fold activity gain in five cycles, and up to 104-fold improvement for Staphylococcus aureus Sortase A. Why it matters: China is shipping AI-plus-robot biocatalyst discovery, not only paper benchmarks. Caveat: results cover selected enzyme tasks, and inventors list pending Chinese patents on the system.

No. 180

Obsidian Security hits a $1.1B mark on $85M to govern AI agents in SaaS apps

Obsidian Security

Palo Alto's Obsidian Security raised $85 million in Series D funding led by Crescent Cove Advisors, with Greylock and Menlo Ventures rejoining, valuing the firm at $1.1 billion and lifting total funding past $200 million. The platform polices non-human identities and AI agents inside third-party apps such as Salesforce, Slack, and GitHub, and is adding native controls for Anthropic's Claude Code and Cowork. Obsidian cites more than 100 customers spending over $100,000 a year and 60 Fortune 500 logos. Why it matters: investors keep funding runtime guardrails as agents touch production SaaS. Caveat: growth metrics come from the company's announcement.

Catch up