OpenAI Reaffirms Zero Data Retention for Frontier Models; Replit Offers GPT-5.6 Luna Free Mode

OpenAI reaffirmed zero data retention for frontier models for API customers who meet eligibility requirements, and previewed Private Safety Processing that does not compromise data privacy. The same channel reported that Replit enabled a free mode powered by GPT-5.6 Luna so ideas can be turned into working software without worrying about token costs. GitHub Copilot added Grok 4.6 and JetBrains enterprise management settings, while Google DeepMind and NVIDIA put robot whole-body intelligence and sign-language AI, plus infrastructure capital and a talent hub, on the official record.
Today's Flow
The axis of this briefing is a path for enterprises to call models without leaving data, and a product message that opens coding and creation tools more widely. OpenAI reaffirmed zero data retention for frontier models for API customers who meet eligibility requirements, and also previewed Private Safety Processing that addresses advanced AI safety without compromising data privacy. Right beside that, Replit said it has turned on a free mode with GPT-5.6 Luna so anyone can turn ideas into working software without worrying about token costs.
On the developer-tools side, GitHub Copilot is rolling out xAI's Grok 4.6 and attached enterprise management settings to the JetBrains plugin. Google DeepMind said Gemini Robotics 2 brings whole-body intelligence to robots, and also posted a piece on putting sign-language AI into users' hands. NVIDIA, calling AI factory computing an investable asset class, introduced a partnership for an independent financial platform designed to mobilize more than $500 billion in third-party capital, and also issued news of Indonesia's first university-based AI technology center. Anthropic posted an article related to Claude text watermarks. Claims of direct application to bio and healthcare are not confirmed in currently published official materials.
Key News
Topic: OpenAI reaffirms zero data retention for frontier models and previews a safety-processing path
OpenAI's official channel said it provides zero data retention for frontier models to API customers who meet eligibility requirements. The materials described this as a reaffirmation, and also said the company is previewing Private Safety Processing that addresses advanced AI safety without compromising data privacy. It reads as an intention to place the option of not leaving call records and safety review on the same privacy premise.
Detailed conditions for who is covered, exceptions, retention periods, whether data is used for training, and industry-by-industry adoption cases do not appear in this evidence as figures or clauses. Assessments that barriers are falling in finance, healthcare, and legal settings are also outside the official wording. The confirmed fact is that zero data retention has been put forward again for the frontier model API, and a privacy-preserving safety-processing path has been previewed.
Topic: Replit widens access to software creation with GPT-5.6 Luna free mode
OpenAI reported that Replit is expanding access to software creation with GPT-5.6 Luna. Replit is introducing a free mode powered by this model, described as letting anyone turn ideas into working software without worrying about token costs.
Usage limits for free mode, differences from paid plans, model unit prices, and revenue sharing are not confirmed in currently published official materials. The interpretation that OpenAI chose ecosystem over its own direct revenue is also outside the evidence. The message that can be pinned this time is a product announcement that it will put GPT-5.6 Luna on free mode and lift creation and development entry costs off token worries.
Topic: GitHub Copilot adds Grok 4.6 and opens enterprise management settings for JetBrains
GitHub said xAI's latest reasoning model, Grok 4.6, is being rolled out in GitHub Copilot. The model is described as designed for agentic coding and complex multi-step workflows. Internal testing is mentioned, but comparison targets and performance figures are not confirmed in the current materials because the public sentence is cut off mid-way.
In the same product family, enterprise management settings were added to the GitHub Copilot JetBrains plugin. Administrators can apply plugin governance, MCP server access, OpenTelemetry, and permission modes consistently across the organization. Model choices are expanding, and a move to centralize plugin permissions and observability has appeared alongside them. This article does not claim that security effects on large organizations have been proven in numbers.
Topic: Google DeepMind highlights whole-body intelligence with Gemini Robotics 2 and also posts sign-language AI
Google DeepMind posted that Gemini Robotics 2 brings whole-body intelligence to robots. The definition of whole-body intelligence, hardware models, benchmarks, and a commercial timeline are not confirmed in currently published official materials. How individual joint control and environmental interaction are integrated also cannot be asserted because the body text is missing.
The same channel also posted a piece on putting sign-language AI into users' hands. Supported sign languages, devices, and launch scope are not in this evidence. It can be read as an announcement to extend on-screen dialogue models to the body and a sign-language interface, but expansion into logistics, autonomous driving, or medical robots is not stated.
Topic: NVIDIA treats AI factory computing as an investment asset and also opens a university hub
NVIDIA announced a partnership with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. It said it is designing an independent financial platform to mobilize more than $500 billion in third-party capital over time to back the build-out of AI infrastructure. It used the phrasing that AI factory computing is becoming an investable asset class. Actual committed amounts, execution schedules, and platform-by-platform shares are not broken out as numbers in this article.
Separately, Indonesia's Ministry of Communication and Digital, Indosat Ooredoo Hutchison, NVIDIA, and Universitas Gadjah Mada said they have opened the UGM Indosat NVIDIA AI Technology Center in Yogyakarta. It is Indonesia's first university-based AI technology center, and the stated purpose is to train local AI talent. There is a mention that it was set up under a national AI center framework, but the following sentences are cut off, so the overall structure of the parent program cannot be confirmed.
Topic: Anthropic posted an article related to Claude text watermarks
Anthropic's official channel posted news on Claude text watermarks. Where the watermark is embedded, who detects it, and which products and plans it applies to are not confirmed in this evidence because the body text is missing. An intent to identify generated text can be read from the title, but there is no basis for stretching that into a regulatory-response or external-tool integration strategy.
AI Γ Bio Points to Watch
Topic: There is no official confirmation that zero data retention is used for bio data
There is no announcement that genomes, patient records, or clinical data will be analyzed on this path. Still, putting a no-data-left-behind option back on the table for eligible API customers and previewing privacy-preserving safety processing is an item organizations that handle sensitive information will check when they weigh the terms. Pharma and hospital adoption, or faster joint research, still need separate contracts and regulatory judgments; this material alone does not mean the standard has changed.
Topic: No allocation plan linking infrastructure capital to bio compute demand has appeared
The more than $500 billion in third-party capital is aimed at AI infrastructure build-out in general. How much would go to protein-structure prediction or molecular simulation is not stated. Indirect effects on compute-expensive life-science research remain a point to watch, but claims that healthcare data centers or the bio-startup investment climate have changed are not confirmed in currently published official materials.
Topic: There is no description that whole-body-intelligence robots lead to lab automation
Gemini Robotics 2 only put forward whole-body intelligence for robots; it did not mention bio robots, cell culture, or manipulating microfluidic devices. Sign-language AI is an accessibility message. A shift into laboratory automation is a topic to watch for future technology transfer, not the announced scope this time.
Sources
- OpenAI: Offering Zero Data Retention for Frontier Models
- OpenAI: Replit Expands Access to Software Creation with GPT-5.6 Luna
- GitHub Changelog: Enterprise Management Settings in GitHub Copilot for JetBrains
- Anthropic: Claude Text Watermark
- NVIDIA: UGM, Indosat, NVIDIA: Indonesia's First University AI Technology Center
- GitHub Changelog: Grok 4.6 Provided in GitHub Copilot
- Google DeepMind: Gemini Robotics 2, Whole-Body Intelligence for Robots
- NVIDIA: AI Factory Computing, an Investable Asset Class
- Google DeepMind: Putting Sign Language AI into Users' Hands
For enterprises to use frontier models, a path that does not leave call data and safety review must sit on the same privacy premise. OpenAI's reaffirmation locks that condition into product language. Replit's free mode and GitHub Copilot's model and permission updates move access and control on the creation and development floor at the same time. Robot whole-body intelligence and AI infrastructure capital show off-screen deployment and fundraising scale, but bio-specific applications are not yet in the official wording.