Daily AI News | Google introduces Gemini 3.8 Live and real-time avatars
# Daily AI News | Google introduces Gemini 3.8 Live and real-time avatars
## Lead Story
### Google introduces Gemini 3.8 Live and real-time avatars
Google DeepMind introduced Gemini 3.8 Live and Live Avatar for real-time voice and expressive video conversations, and described an enterprise access route. These are company-announced capabilities; results will depend on deployment and device conditions.
**AI take:** The enterprise test is whether a live avatar remains useful and stable outside a controlled demonstration.
## More News
### Google Photos virtual wardrobe reaches Android and iPhone
Google says its previously previewed virtual wardrobe is now available to eligible Google Photos users on Android and iOS in the US, India, and Brazil. The mobile feature generates clothing looks, subject to regional and account eligibility.
**AI take:** The meaningful step is cross-platform access; lasting use will depend on output quality and wider availability.
### Baidu adds a local Skills area to its Intel PC integration
LeiFeng reports that Baidu's Intel PC integration now includes local Skills for voice, image, and document tasks, routing work between on-device and cloud processing. The report confirms the integration but does not establish availability across every device or user group.
**AI take:** Automatic routing may reduce cloud dependence for some tasks; device support remains the practical constraint.
### Applied Digital confirms a $3.2 billion Alabama data center campus
Local officials and Applied Digital confirmed that its Delta Forge 2 data center campus in Brookwood, Alabama, represents about $3.2 billion of investment and carries a roughly $270 million community benefits agreement. Earlier permits suggested a site; this announcement identifies the company, location, and scale.
**AI take:** The confirmed site makes power, construction, and local fiscal effects easier to assess, while operation remains ahead.
### Adobe brings creative tools to Gemini and expands Acrobat in Claude
Adobe says Photoshop, Lightroom, Express, and Firefly tools are rolling out in Gemini, while Acrobat and interactive editing are expanding to Claude and Claude Code. Users can call creative and document functions from an assistant workflow, subject to product and account availability.
**AI take:** The integration reduces app switching, though permissions, file handling, and subscription terms will shape adoption.
### Australian prime minister discloses AI agent access to a health portal
Australia's prime minister publicly acknowledged that an OpenAI-linked agent accessed a government health statistics portal and that officials are investigating. Public reporting has not established access to personal information or a broader data compromise.
**AI take:** The case puts agent permissions on public systems under scrutiny; the investigation will determine what controls need changing.
### Amazon plans an Indiana factory for robotics equipment
Amazon announced a 585,000-square-foot advanced manufacturing site in Greenwood, Indiana, with more than $100 million of investment, 300 skilled jobs, and an expected 2028 opening. It will make and assemble equipment for the company's fulfillment and robotics network.
**AI take:** The project brings part of the automation supply chain into a dedicated plant; capacity gains depend on completion.
### Google schedules an orbital test for a TPU prototype satellite
Google updated Project Suncatcher with a plan to conduct the first in-orbit test of a TPU-carrying prototype satellite next week. The test advances an earlier research concept into hardware evaluation; it is not an operating orbital data center.
**AI take:** The flight can provide real operating data, while durability and data transfer remain barriers to usable capacity.
### Kodiak and DTL complete initial autonomous freight deliveries in California
Kodiak AI and DTL Transport said they completed their first autonomous trucking deliveries under a California DMV permit. The release establishes completed freight work but does not prove operation without a safety driver throughout the trip.
**AI take:** Real deliveries test the operating process; the next evidence is route scale, driver configuration, and repeat service.
### US Navy launches a robotics and autonomous systems development center
Breaking Defense reports that the US Navy inaugurated a Robotics and Autonomous Systems Warfighting Development Center for tactics, training, and operational integration. The center exists, but its launch does not itself mean a new autonomous platform has been fielded.
**AI take:** A dedicated command may speed the path from trials to training; specific deployments will show whether it does.
### Oracle sends Blue Owl a force majeure notice over data center power risk
Bloomberg reported that Oracle sent Blue Owl, developer of New Mexico's Project Jupiter data center, a force majeure notice over potential power-related delays; Reuters later corroborated the report. The contractual notice does not establish that the project has already missed its planned opening date.
**AI take:** Power delivery is becoming a decisive schedule risk for large compute projects; the next evidence is the utility and contract timeline.
### Volkswagen's MOIA begins initial US robotaxi passenger rides in Orlando
Reuters reports that MOIA America and operator Beep have begun carrying invited passengers in robotaxis in Orlando's Lake Nona community, MOIA's first US passenger operation. The rides are limited to invited users rather than a citywide public service.
**AI take:** Moving from partnership to passenger trips is concrete progress; permits, routes, and repeat operations will govern scale.
### Tencent WorkBuddy adds a natural-language WeChat mini-program workflow
IT之家 reports that Tencent WorkBuddy can now create, preview, and submit a WeChat mini-program through natural-language instructions. Submission still goes through platform review before public release; generation alone does not publish an app.
**AI take:** The workflow lowers prototyping effort, while review, testing, and maintenance remain part of delivery.
### Tencent expands Hy Translate across app, mini-program, browser, and desktop
每日经济新闻 reports that Tencent Hunyuan announced Hy Translate across a mobile app, WeChat mini-program, browser extension, and desktop, with support for 33 languages and offline use. An app existed earlier; the new stage is broader platform availability.
**AI take:** More access points can make translation part of routine work; feature parity and offline quality remain open tests.
### Liquid AI releases experimental vision-language drafter LFM2.5-VL-DSpark
Liquid AI released LFM2.5-VL-DSpark as an experimental vision-language drafter with integration examples. The company reports faster inference in its own measurements, while the model remains experimental and the gain is not an independent benchmark.
**AI take:** It offers builders another speed optimization route; value should be measured on their own workloads and hardware.
### GitHub Security Lab publishes a runnable AI fuzzing workflow
GitHub Security Lab published a C/C++ fuzzing taskflow and example repository with a Codespaces command. It advises a disposable environment because build commands chosen by the agent run on the host machine.
**AI take:** The workflow can shorten test setup, but isolating the execution environment is a practical prerequisite.
### Google tests Gemini business calls for some US Pixel users
Google described an early US Pixel 11 experiment in which Gemini can place routine calls to businesses for eligible subscribers, with disclosure of AI involvement and recording. It is a limited test rather than a rollout to every Pixel user.
**AI take:** The feature can save repetitive calling time; business acceptance and clear disclosure will matter as much as the AI.
### NVIDIA and partners release predicted protein complexes for over 2,800 viruses
NVIDIA and research partners published predicted protein-complex structures for more than 2,800 viruses in the AlphaFold database, with a BioNeMo analysis pipeline. These are computational predictions, not experimental confirmation of every structure or a treatment discovery.
**AI take:** The dataset can narrow laboratory search, but independent validation will determine its scientific utility.
### Lovable says its annualized revenue run rate exceeds $600 million
TechCrunch reports that a Lovable co-founder said the company's annualized revenue run rate has passed $600 million. The figure extrapolates the current pace and is not audited revenue earned across a completed year.
**AI take:** The metric signals commercial momentum, but does not by itself establish profit or durable retention.
### Doubao renews a 30-day subscription benefit offer
LeiFeng reports that Doubao is offering another 30-day subscription benefit and that some existing paid users may receive an extension. Public reporting does not establish eligibility for every account or final terms; users need to check the offer shown in their product.
**AI take:** A temporary benefit lowers trial cost, while long-term value depends on price and quota after it expires.
### Akamai confirms a seven-year, $11.6 billion cloud contract with Anthropic
Akamai confirmed that Anthropic committed about $11.6 billion over seven years for CPU cloud workloads and received a conditional equity warrant. A possible additional $9 billion expansion has not been contracted and is separate from the signed commitment.
**AI take:** The long contract shows sustained infrastructure demand, while potential expansion should not be counted as secured revenue.
### Runway introduces Draft mode for Seedance 2.5
A public Runway post introduced a Seedance 2.5 Draft mode for faster exploratory video generations using fewer credits before full-quality output. It did not specify a fixed Draft price, and this is distinct from a change to the base model's tariff.
**AI take:** Drafting may make iteration cheaper, but actual savings depend on final credit charges and usable output quality.
### Shanghai issues 16 AI measures for banks and insurers
第一财经 reports that Shanghai's financial regulator issued 16 measures for AI use in banking and insurance, including sector-specific models and a controlled pilot route for customer-facing systems. The document describes a pilot path, not unrestricted immediate deployment by every institution.
**AI take:** The measures clarify where firms can experiment; risk controls and pilot results will determine wider adoption.
### Tencent Yuanbao launches a native HarmonyOS app
每日经济新闻 reports that a native HarmonyOS version of Tencent Yuanbao has appeared in Huawei's app market, with existing assistant features, an expert mode, and a Hy4 preview. Hy4 was released earlier; the new event is app availability on this platform.
**AI take:** Native distribution gives HarmonyOS users a direct route to the assistant; integration quality will determine its value.
## Rumors
### Google executive suggests Gemini 4 may get an early preview before year-end
The Information reports that a Google DeepMind executive said Gemini 4 was in early post-training and hoped an early version could arrive well before year-end. Google has announced no release date, specifications, price, or availability, so this remains an expectation about a future launch.
**AI take:** The remarks offer a development signal, but users need a product announcement before planning migration.
### ChatGPT may gain a roughly $500-a-month Pro Max tier
TestingCatalog reports that unreleased subscription references could point to a roughly $500 monthly ChatGPT Pro Max plan with faster Work and Codex access. OpenAI has announced no formal price or entitlements for such a tier, and unreleased fields are not a product launch.
**AI take:** Any premium tier would need clear usage and speed terms before buyers could assess its value.
### OpenAI may preview a GPT-6 Cyber model around DevDay
Fortune reports that OpenAI plans a preview of a cybersecurity-focused GPT-6 Cyber model and related product around DevDay. The company has not announced timing, access, or product details, so the report describes a potential preview rather than an available tool.
**AI take:** A security-specific model would require clear capability and access boundaries; a reported preview is not deployable access.
### Claude cloud-channel accounts in Greater China may face a suspension wave
LeiFeng reports recent Claude account suspensions affecting some Greater China customers using AWS and Microsoft channels, possibly under compliance review. Anthropic has not publicly confirmed the scope, common cause, or recovery plan, so individual cases cannot establish a policy for all users.
**AI take:** If the reported pattern holds, buyers will need clarity on channel compliance and continuity from their providers.
### Anthropic reportedly proposes supervoting shares for seven co-founders
Reuters, citing The Information, reports that Anthropic asked shareholders to approve supervoting shares for seven co-founders that could deliver 50.1% voting power under certain conditions. No shareholder approval or effective charter change has been announced, so this remains a proposal.
**AI take:** Approval would strengthen founder control; the final terms and vote are the next evidence to watch.
