Artificial Curiosity Labs
Go back

AI News Digest

AI News Digest — 2026-08-15

Top Stories

The Week’s 10 Biggest Funding Rounds: Data, Neolab, AI Infrastructure, Defense And AI Coding Lead (Crunchbase News)

The week’s biggest rounds were concentrated in data, AI infrastructure, defense and AI coding, with Databricks returning for another $5 billion just eight months after its last raise — a sign that AI funding remains hot.

The builder’s guide to GPT‑5.6 (OpenAI)

OpenAI’s builder guide for GPT‑5.6 outlines new model-selection tools and Responses API enhancements that enable developers to build agents that run faster and at lower cost.

GLM-5.3 Arrives Hours After Tang Jie’s ‘sooooooon’ — Zhipu’s Coding and Security Model Doubles SWE-Marathon, Tops CyberGym (Pandaily)

Zhipu’s GLM-5.3, a coding- and security-focused model, landed hours after Tang Jie’s “sooooooon” teaser, doubling SWE-Marathon and topping CyberGym — a fast follow that sharpens open-model competition in agentic software engineering.

DeepSeek V4 Pro API Update Adds Responses API Support (TechNode)

DeepSeek’s quiet integration of its V4 Pro model with Responses API support, a 1M-token context window, and aggressively low pricing represents a direct competitive thrust at established frontier model providers by combining top-tier capability with a developer-friendly interface and cost structure designed to win API market share.

Apple trained its own AI model for China with help from Alibaba (The Verge)

Apple’s collaboration with Alibaba to train a domestic AI model for China underscores how geopolitical friction between the US and China is pushing tech firms to tailor artificial intelligence locally and lean on critical cross-border alliances.

Foundation Models

What moved

GLM-5.3 Arrives Hours After Tang Jie’s ‘sooooooon’ — Zhipu’s Coding and Security Model Doubles SWE-Marathon, Tops CyberGym (Pandaily)

Zhipu’s coding- and security-focused GLM-5.3 landed just hours after Tang Jie’s “sooooooon” teaser, doubling its SWE-Marathon result and taking the top spot on CyberGym, which sharpens the open-model race for agentic software engineering by pairing rapid release momentum with benchmark-leading performance in coding and security tasks.

Meta’s ‘open’ AI, and a $250M deal gone very wrong (TechCrunch)

Meta’s simultaneous release of the open-weight Glimmer model alongside the API-gated Muse Spark underscores a deliberate bifurcation in its AI strategy, reigniting the industry debate over whether open models truly democratize access or merely serve as loss leaders for proprietary ecosystems.

OpenAI and Anthropic in price war as Chinese AI rivals gain ground (Ars Technica)

The aggressive price cuts from OpenAI and Anthropic, driven by mounting pressure from rapidly advancing Chinese AI labs, mark a shift from a contest of pure capability to one of cost efficiency, threatening the premium pricing that has funded frontier model development. As Chinese rivals close the performance gap, the resulting margin compression could force Western firms to sacrifice investment in next-generation breakthroughs, potentially yielding the innovation lead to competitors that operate with lower structural costs and state-backed scale.

Also tracking

Robotics & Physical AI

What moved

X Square Robot Set Its Benchmark at Figure AI’s Number — Then Beat It by 45% (Pandaily)

Shenzhen’s X Square Robot directly challenged Figure AI’s benchmark by first matching its sustained parcel-sorting rate, then surpassing it by 45% during an unedited, hour-long test, injecting a concrete performance metric into the competitive humanoid robotics race.

BYD-Backed Deep-Sea Robot Maker Shenhai Zhiren Raises 500 Million Yuan, Debuts Underwater Agent Model SEAgent 1.0 (Pandaily)

Shenhai Zhiren, a deep-sea robotics company backed by BYD, has secured 500 million yuan in Series A funding and unveiled its underwater agent model, SEAgent 1.0, demonstrating simultaneous capital and technical momentum in the niche of physical AI for subsea operations.

Gemini Robotics ER 2: powering robotics with video understanding, task orchestration, and multi-robot collaboration (deepmind.google)

By integrating video understanding, task orchestration, and multi-robot collaboration, Gemini Robotics ER 2 directly tackles the perception, planning, and coordination hurdles that have kept robots confined to narrow settings, marking a significant step toward deployable real-world systems.

Also tracking

Agents & Tooling

What moved

AI Starts Designing Chips: EDA Giants Synopsys, Cadence, and Siemens EDA Open the Agentic AI War — Chinese Vendors See a Golden Window (Pandaily)

The race to automate chip design is now shifting from assistive AI to autonomous agents: at DAC 2026, Synopsys, Cadence, and Siemens EDA all unveiled agentic AI chip-design strategies alongside Chinese challengers, with Kimi’s K3 autonomously designing a chip for 48 hours. That matters because agentic AI is moving directly into semiconductor implementation workflows, intensifying competition between entrenched EDA incumbents and Chinese vendors looking for a window to close the gap.

Weak API controls are one of the biggest threats in the agentic AI era (SiliconAngle)

As agentic AI systems gain autonomy inside enterprise workflows, poorly governed APIs turn into a critical and underappreciated attack surface, making weak API controls one of the biggest threats in this new era.

75% of developers I surveyed prefer Claude Code - here’s why they choose it over Codex (ZDNET)

A recent developer survey found 75% of respondents now favor Anthropic’s Claude Code over OpenAI’s Codex for daily coding workflows, pointing to a significant realignment in AI coding tool adoption that could influence enterprise toolchain decisions.

Also tracking

Voice & Speech

What moved

China’s ‘Thinking Machines’: VUI Labs’ Luna-TTS Tops the Global TTS Arena, Beating ElevenLabs and MiniMax (Pandaily)

VUI Labs’ Luna-TTS ranking first on Hugging Face’s TTS Arena and third on Artificial Analysis’ Speech Arena, ahead of ElevenLabs and Google, marks a notable shift in open voice-model leadership and raises the competitive bar for synthetic speech quality.

Gemini voice calling on Android Auto keeps failing me - and Google has until September to fix it (ZDNET)

Gemini’s persistent voice-calling failures on Android Auto, just as Google prepares to retire Assistant, expose a critical reliability gap in its AI transition — with a September deadline looming, the company must prove its new assistant can handle essential, safety-critical driving functions as seamlessly as the one it’s replacing.

We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control (deepmind.google)

DeepMind’s Lyria 3.5 integration into Google Flow Music advances the generation model’s musicality, lyric coherence, vocal quality, and user creative control, marking a step forward in AI‑generated audio fidelity and artistic flexibility.

Policy & Safety / Other

What moved

Google will now allow users to remove visible watermark from its AI generations (TechCrunch)

Google is recalibrating the line between disclosure and user control, allowing creators to remove visible SynthID watermarks from images and expanded audio outputs, while invisible digital provenance signals are maintained. The shift responds to user pushback over aesthetics cited by TechCrunch, but it risks normalizing the pristine distribution of deepfakes and complicates public literacy about synthetic media origins even as the company emphasizes its continued partnership with C2PA standards.

Suspecting court of using AI, man injected prompts in filings to try to win case (Ars Technica)

A litigant’s attempt to manipulate a suspected AI-driven court by embedding prompts in filings exposes the mounting risks of chatbot misuse in pro se litigation, drawing fresh judicial warnings about unverified AI-generated legal arguments.

People Are ‘Marrying’ Chatbots. These Lawmakers Want to Stop Them (Wired)

State lawmakers are moving to preemptively block legal recognition of human-AI marriages, reflecting growing unease over the deepening emotional bonds people are forming with chatbots and setting an early regulatory boundary for AI companionship.

Also tracking

Infra & Compute

What moved

Chipmaking equipment supplier Lam Research to spend $3B+ on new labs (SiliconAngle)

Lam Research’s planned $3 billion-plus investment in new labs is a direct bet on sustained AI-driven chip demand, expanding the tooling capacity that underpins advanced logic and memory fabs.

Kog is going deeper to squeeze more inference out of GPUs (TechCrunch)

French startup Kog’s push into deeper GPU inference optimization matters because it challenges the assumption that GPUs are poorly suited for agentic workloads, signaling a potential shift in how such workloads are deployed.

Hyperscalers might regret embracing natural gas if new forecast proves correct (TechCrunch)

Hyperscalers betting on natural gas to power their expanding AI data centers now face a stark economic threat: a new forecast warns that natural gas prices could triple, which would sharply undercut the cost advantages they sought and pressure the profitability of their energy-intensive operations.

Also tracking

Applications

What moved

Google Meet can take notes for your in-person meetings now - here’s how it works (ZDNet)

Google Meet’s Gemini-powered note-taking, previously limited to virtual calls, now captures in-person meeting audio via a laptop microphone, providing real-time transcriptions and summaries. This blurs the boundary between physical and digital collaboration workflows by automatically documenting spoken decisions and action items, which reduces the risk of lost context when teams have hybrid interactions.

ChatGPT’s new Computer History tracks your Mac activity to create a timeline - but should you let it? (ZDNET)

ChatGPT’s Mac app introduces a Computer History feature that monitors activity across applications and websites to construct a chronological timeline, directly surfacing concerns about the extent of data collection and its safeguarding.

I used OpenFactory to build my own Linux distro overnight - this AI tool is going to be big (ZDNET)

Generative AI is expanding from code completion into full operating-system assembly: ZDNET reports that OpenFactory let a user build a custom Linux distribution overnight, signaling that complex system configuration and distro creation may become accessible to far more people through automation.

Also tracking

Funding & Capital

What moved

The Week’s 10 Biggest Funding Rounds: Data, Neolab, AI Infrastructure, Defense And AI Coding Lead (Crunchbase News)

Databricks is back for another $5 billion round just eight months after its last raise, landing alongside large new deals in AI infrastructure, defense and AI coding — a signal that AI funding remains hot across multiple layers of the stack.

Accelerating the frontiers of scientific discovery: Google’s $40M commitment to the Genesis Mission (deepmind.google)

Google’s $40 million commitment to the Genesis Mission, delivered as AI tokens and credits, matters because it directly lowers compute barriers for researchers, accelerating the pace of AI-driven scientific discovery across disciplines.