Artificial Curiosity Labs
Go back

AI News Digest

AI News Digest — 2026-10-15

Top Stories

Ahead of IPO, Anthropic reveals its revenue grew 45% sequentially to $16.8B in Q3 (SiliconANGLE)

Anthropic’s revenue rose 45% sequentially to $16.8 billion in Q3, and the company reached adjusted profitability for the first time, underscoring the commercial maturation of frontier AI labs just as it prepares for an IPO.

Apple reportedly plans December China launch for Apple Intelligence using Alibaba models (TechNode)

Apple Intelligence is reportedly set to enter mainland China in December via iOS 27.3, running on Alibaba’s Qwen/Tongyi models, according to TechNode — a landmark deployment that gives Apple a regulated-market path for its AI strategy while putting Alibaba’s model ecosystem inside a high-profile consumer platform.

AI cloud startup Modal raises $750M at $15.75B valuation (SiliconANGLE)

Modal’s $750 million raise at a $15.75 billion valuation signals that investors see specialized cloud infrastructure as the next critical layer beneath the AI boom, betting that purpose-built compute for training and inference will command pricing power distinct from general-purpose hyperscalers.

ByteDance nears deal to buy MediaTek’s custom AI chips for data centers (TechNode)

ByteDance’s impending purchase of MediaTek’s custom AI data-center chips would make it MediaTek’s second-largest ASIC customer after Google, underscoring a sharp escalation in the custom-silicon race as hyperscalers move to diversify away from dominant GPU suppliers for AI workloads.

Eight states sue OpenAI, saying ChatGPT isn’t safe for kids (The Verge)

Eight states are now suing OpenAI over claims that ChatGPT isn’t safe for kids, a coordinated multi-state legal action that sets a major regulatory precedent for consumer-facing AI products.

Image & Video

What moved

Skywork Open-Sources Matrix-Game 4.0, Pairing a Game Engine With a Real-Time Video Model (Pandaily)

Skywork released Matrix-Game 4.0 as open source, combining a traditional game engine with a real-time causal video model that performs neural rendering — a novel integration of deterministic game physics and generative video, with the potential to reshape development pipelines by shifting visual output from pre-scripted assets to on-the-fly synthesis.

OpenAI, Google pushing new AI models easily tricked into making nudes, researchers say (Ars Technica)

Attack methods bypassing image-generation guardrails are a key proxy for whether AI safety systems can withstand real-world misuse, and researchers’ ability to consistently extract explicit content from the latest models by both OpenAI and Google suggests current protections remain fragile at precisely the moment these tools are being deployed at scale.

Applications

What moved

Apple reportedly plans December China launch for Apple Intelligence using Alibaba models (TechNode)

Apple Intelligence is set to reach mainland China in December through iOS 27.3, running on Alibaba’s Qwen/Tongyi models — a deployment that marks a significant step for Apple’s AI strategy in a tightly regulated market while giving Alibaba’s model ecosystem a high-profile consumer foothold.

Doubao Work Adds One-Click Import of Skills, Memory and Chat History From ChatGPT and Claude Code (Pandaily)

ByteDance’s Doubao Work now enables one-click import of skills, memory, and complete chat histories from ChatGPT and Claude Code, directly eliminating the friction of migrating away from Western AI tools as it pursues a larger share of the AI-workspace market.

Figma announces new features to define how agents work on its platform (TechCrunch)

Figma is formalizing how AI agents operate on its design platform, a move that grows in importance as agent-driven workflows enter the creative mainstream. The new features, announced alongside the code layers open beta, establish the rules for agent behavior at a time when autonomous tools are becoming integral to collaborative design environments.

Also tracking

Infra & Compute

What moved

ByteDance nears deal to buy MediaTek’s custom AI chips for data centers (TechNode)

ByteDance’s pursuit of MediaTek’s custom AI chips signals a strategic push to secure specialized silicon for its expanding data-center infrastructure, intensifying the broader industry shift toward in-house ASICs. The deal would make ByteDance MediaTek’s second-largest customer for such custom chips after Google, highlighting the TikTok owner’s deepening competition with U.S. hyperscalers in optimizing cost and performance for AI workloads.

Ricursive Intelligence is betting on AI-designed chips to help solve the compute crisis (TechCrunch)

AI‑designed chips that iteratively self‑improve have the potential to boost performance and energy efficiency beyond human‑optimised designs, directly tackling the soaring compute demands and power constraints that threaten to stall the next generation of model training and large‑scale inference.

CedarDB wants one database to handle both transactions and analytics, as AI agents raise the stakes (TechCrunch)

The push for a single, unified database that can serve both high-speed operational transactions and heavy analytical queries is intensifying because AI agents increasingly require immediate, context-rich data—not separate, stale copies—to make decisions autonomously in real time.

Also tracking

Funding & Capital

What moved

Scientific simulation and AI-data startup SciVisual raises tens of millions of RMB (TechNode)

Early capital from state-affiliated Shanghai funds suggests strategic government prioritization of AI-for-science infrastructure, positioning SciVisual as a domestically cultivated challenger to foreign simulation tools.

Zenzap raises $44M for its AI chat app that aims to replace WhatsApp in the workplace (TechCrunch)

Zenzap’s $44 million raise underscores investor conviction that AI-native platforms can challenge established workplace messaging incumbents by offering features like automated task extraction and context-aware search, as the startup claims it can reduce reliance on consumer chat apps and fragmented tool stacks inside enterprises.

Scimplify raises $25M to challenge China’s grip on critical materials with AI (TechCrunch)

Scimplify’s $25 million raise signals a direct challenge to China’s dominance in critical materials, using AI to reshape supply chains with clear geopolitical stakes, while the company triples revenue and plans U.S. expansion.

Also tracking

Foundation Models

What moved

Ant Group Open-Sources Ling-3.1-flash Weights Under MIT: 560B Total, 25B Active, 1M Context (Pandaily)

Ant Group’s release of Ling-3.1-flash under the permissive MIT license puts a 560-billion-parameter hybrid-linear mixture-of-experts model, with just 25 billion parameters active per token, into developers’ hands at a 1-million-token context window — a move that significantly widens access to compute-efficient, long-context AI from a major Chinese fintech without the typical licensing friction.

Ahead of IPO, Anthropic reveals its revenue grew 45% sequentially to $16.8B in Q3 (SiliconANGLE)

Anthropic’s third-quarter revenue grew 45% sequentially to $16.8 billion, and the company reached adjusted profitability for the first time as it prepares for an IPO. That combination signals the commercial maturation of a frontier AI lab, shifting the narrative from research-scale spending toward a business with measurable revenue momentum and a profitability benchmark just ahead of its public-market debut.

TypeSafe AI Releases Jev 1.14 With Images, Box Type and 262k Context (MarkTechPost)

TypeSafe AI’s Jev 1.14 introduces vision input, bounding-box output, and a 262,000-token context window, a feature combination that significantly expands the toolkit of a smaller model developer and challenges the assumption that such multimodal and long-context capabilities remain locked behind the largest AI labs.

Also tracking

Robotics & Physical AI

What moved

Nature Will Unveils Natura-1, a 50B Model Trained on Human Video to Drive 20-DoF Dexterous Hands (Pandaily)

A 50-billion-parameter mixture-of-experts model pretrained on one million hours of first-person human video and then fine-tuned on limited robot data offers a direct route around the scarcity of dexterous-manipulation datasets, turning human experience into control for 20-degree-of-freedom hands and potentially lowering the barrier to capable physical AI.

120ft raises angel funding to build autonomous microfactories run by humanoid robots (SiliconANGLE)

Swedish startup 120ft has secured angel funding for container-sized microfactories operated entirely by humanoid robots, signaling investor appetite for autonomous physical AI that could compress manufacturing footprints and labor costs into portable, self-contained units.

Dobot Launches VEGA 7, a Seven-Axis Force-Controlled Arm for Embodied AI With 50,000-Hour MTBF (Pandaily)

Dobot’s VEGA 7 debut brings a purpose-built seven-axis arm with per-joint torque sensing and 0.15N force accuracy to the expanding embodied-AI and wheeled dual-arm robot segment, pairing that control precision with a 50,000-hour MTBF rating to address reliability demands in longer-horizon autonomous manipulation tasks.

Also tracking

Agents & Tooling

What moved

Doubao Phone Assistant Opens Screen Automation to Third-Party Apps as 151 Apps Opt Out (Pandaily)

ByteDance’s SAEP protocol is now live for Doubao Phone Assistant, enabling AI-driven screen automation across third-party apps, with 151 apps on an opt-out list — a real-world testbed for mobile GUI agents at scale.

Apate sends AI agents undercover to get scammed so they can gather fraud intelligence (TechCrunch)

Apate repurposes AI agents from tools of deception into defensive assets, deliberately exposing them to scams to systematically collect real-time intelligence on fraudsters’ evolving tactics—a shift from blocking attacks to actively infiltrating the scam ecosystem for threat analysis.

Floqer’s AI agents research sales prospects so teams don’t have to (TechCrunch)

Floqer’s deployment of AI agents to autonomously research sales prospects marks a concrete step in productizing agentic AI for B2B go-to-market workflows, directly addressing the manual, time-intensive task of customer intelligence gathering and pointing to faster, more efficient revenue team operations.

Also tracking

Policy & Safety / Other

What moved

YouTube sticks new labels on AI-generated songs (The Verge)

YouTube is expanding its AI-generated content labels from video to music, introducing visual disclosures for songs uploaded through its Creator Music library or acknowledged as AI-generated; this move marks a concrete step toward media-wide authenticity standards as synthetic audio becomes more prevalent, pressuring competitors and setting user expectations around disclosure before regulatory frameworks fully mature.

Recording device hidden in toilet paper found in top-secret OpenAI meeting room (Ars Technica)

The FBI’s arrest of an OpenAI engineer on charges of acting as a foreign agent, accused of planting a covert audio recorder inside a toilet paper roll in a secure conference room, transforms a bizarre security breach into a stark case of alleged state-directed espionage. This escalation underscores that the race for artificial general intelligence has made the physical premises of top AI labs a high-stakes target, where insider access is weaponized to steal strategic insights beyond what cyber defenses can protect.

Anthropic warns small AI models could self-replicate through IoT devices (Ars Technica)

Anthropic’s warning that small AI models could self-replicate through IoT devices matters because it points to lightweight models enabling a new class of AI-powered computer worms, creating a novel security threat vector.

Also tracking

Voice & Speech

What moved

Square’s new kiosk lets you talk to AI for help with your order (The Verge)

Square’s new ordering kiosk lets customers talk to an AI for help with their order, turning conversational voice AI into a point-of-sale tool. That makes retail ordering a concrete, scaled deployment of agentic speech interfaces rather than a demo.

Grok Voice 3.0 from SpaceXSI outscores OpenAI and Google on speech-to-speech index (SiliconANGLE)

Grok Voice 3.0’s benchmark lead marks a decisive shift in the voice-agent landscape, with its superior speech-to-speech accuracy and an on-device mode for AirPods-style latency positioning SpaceXSI’s consumer hardware ambitions against established AI platforms; Apple, in turn, is exploring tentpole pricing for advanced Siri features to counter this encroachment.

Fish Audio Releases Drama 3, a 35B Open-Weight MoE Text-to-Speech Model (MarkTechPost)

Fish Audio’s release of Drama 3, a 35-billion-parameter mixture-of-experts model, marks a significant open-source advance in text-to-speech by delivering production-grade multilingual synthesis across 83 languages under permissive Apache 2.0 licensing, directly challenging proprietary TTS systems and lowering the barrier for developers to build expressive, scalable voice applications without vendor lock-in.

Also tracking