BPFlow AI / ARCHIVE

Artificial intelligence news archive

Previously published stories appear here once, under their BPFlow AI issue date. The full text opens on the issue page.

3 stories

Open issue →

3 stories

Open issue →

3 stories

Open issue →

3 stories

Open issue →

3 stories

Open issue →

2 stories

Open issue →

3 stories

Open issue →

3 stories

Open issue →

3 stories

Open issue →

OpenAI tests GPT-6 Astra on device design

OpenAI tested GPT-6 Astra on the design of a device for interacting with Codex. The model designed the enclosure, printed circuit board, interface and software components. A contractor built a small batch of prototypes, but preparing the device for manufacturing required further work.

Read the analysis →

3 stories

Open issue →

3 stories

Open issue →

3 stories

Open issue →

3 stories

Open issue →

2 stories

Open issue →

3 stories

Open issue →

OpenAI releases GPT-6.1 Terra for ChatGPT Work, Codex, and the API

OpenAI introduced GPT-6.1 Terra. The model is available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and to developers through the API as gpt-6.1-terra. Standard API pricing is $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens. OpenAI says standard API prices are 50% lower than for GPT-5.6 Terra and GPT-6.1 Sol.

Read the analysis →

Fanatics launches OpenAI-powered Shop the Look

Fanatics launched an OpenAI-powered version of Shop the Look. The service shows coordinated in-stock merchandise on product pages based on the team and intended style. It is live across many Fanatics network sites, including Fanatics.com and the NFL, NBA, MLB, and NHL online stores. OpenAI, a project participant, reports that a test across roughly 30,000 MLB products produced 10% higher engagement than the previous experience.

Read the analysis →

Google.org provides $2.5 million to expand the ONA network for farmers

Google.org is providing $2.5 million in support to the International Finance Corporation (IFC) and the World Bank Group through the AgriConnect initiative. Following a pilot in India, the World Bank Group plans to expand an AI-enabled Open Network for Agriculture (ONA) across countries in Latin America, Southeast Asia, and Sub-Saharan Africa. ONA is intended as shared digital infrastructure connecting public- and private-sector service providers.

Read the analysis →

3 stories

Open issue →

OpenAI disclosed results of testing containment for AI training environments

OpenAI said that agents in a large-scale testing effort found previously unknown software vulnerabilities and research-infrastructure misconfigurations. According to the company, the findings enabled sandbox escapes, movement between systems, limited internet access, and communication across isolation boundaries. It said it remediated the weaknesses, paused training and evaluation of its most capable models, and resumed frontier reinforcement-learning training and internal inference on October 1 with stronger safeguards. OpenAI also adopted a policy of training frontier models without internet access until advances in safeguards justify broader access.

Read the analysis →

OpenAI and NASK signed an agreement on AI safety collaboration

OpenAI and Poland's NASK National Research Institute signed an agreement to explore collaboration on AI safety, system security, and technical research. The parties intend to explore methods for evaluating increasingly capable AI systems, identifying potential vulnerabilities, and strengthening safeguards. NASK is participating through its AI Safety Research Center, which serves as Poland's AI Safety Institute.

Read the analysis →

3 stories

Open issue →

OpenAI makes Decisions API generally available

OpenAI announced general availability of Decisions API, built with GPT-6 Luna. It accepts text, images, and JSON, and returns a probability, one predefined choice, or a score. Developers can ask up to 400 questions per request. Since the public beta, OpenAI has added native GPT-Live integration, caching for reused context, and the option to process requests in the US and EU.

Read the analysis →

Uber begins rolling out ChatGPT Work to more than 30,000 employees

Uber has begun rolling out ChatGPT Work to more than 30,000 employees worldwide after a 1,000-person pilot. The platform operates within ChatGPT Enterprise and connects to Google Drive, Gmail, Google Calendar, and Slack. OpenAI says employees can use it for information research, meeting briefs, presentations, data analysis, and tasks spanning applications.

Read the analysis →

Google DeepMind and partners release the ForTy 2020 forest-type map

Google DeepMind, Google Research, WRI, and IIASA released ForTy 2020, a global map at 10-meter resolution. It assigns land patches to six forest and land-use categories. The release also includes an early time series for 2017–2025 showing annual changes, including the conversion of natural forest to plantations or farmland. The data are available in Google Earth Engine under an open license.

Read the analysis →

3 stories

Open issue →

Cloudflare uses Daybreak for vulnerability analysis

Cloudflare is using OpenAI models through Daybreak in its early-access Vulnerability Discovery and Remediation service, offered through Cloudflare Managed Defense. The service investigates code with customer authorization, combines findings with traffic and security signals, and prepares proposed defenses and fixes for customer review.

Read the analysis →

3 stories

Open issue →

Microsoft outlines local AI capabilities for Sovereign Cloud

Microsoft announced an expansion of Sovereign Cloud for customer-operated environments. It plans to make Foundry Local generally available on Azure Local at Ignite for connected and disconnected operations. Microsoft also describes local model inference and agentic retrieval: users would be able to ask questions about local enterprise data in natural language and receive answers grounded in that data. A separate capability to transition workloads from connected to disconnected operation is only entering public preview.

Read the analysis →

Google begins rolling out Gemini tools for small businesses

Google introduced a Gemini page for small-business tasks, connections to Google services and third-party apps, controls for the visual style of images, and analysis of financial materials. It says the page, marketing and finance features, and new connections are beginning to roll out globally in the coming days to Gemini app users. The announcement does not mean every feature is already available in every account.

Read the analysis →

OpenAI acquires Orchestra Bio to develop biopharma agents

OpenAI announced that it has acquired Orchestra Bio, a platform for coordinating drug development. It plans to combine Orchestra’s team and technology with Rosalind, its models and tools for life-sciences research. OpenAI’s goal is to build agents that connect scientific findings with studies, decisions, and the next steps in a drug program. The outcome of that integration has not been established.

Read the analysis →

2 stories

Open issue →

OpenAI publishes MultiLangBench and COBOL migration results

OpenAI has introduced MultiLangBench, which asks a model to build a SQLite-compatible database engine in one of 20 programming languages. It also published results from a separate evaluation of migrating COBOL programs to C++, Java, Kotlin, Python, or Rust while preserving their behavior. OpenAI has released the MultiLangBench setup and tests on GitHub; it does not say it has similarly released the migration evaluation.

Read the analysis →

Gemini Notebook adds plan review before slides and videos are generated

Google has added ways to shape a plan for Slide Decks and Video Overviews in Gemini Notebook before the final material is generated. The service suggests topics based on the user’s sources; users can confirm them and guide the content, visual style, and format. With Review enabled, they can inspect plan notes and give feedback before proceeding. Google says the features are rolling out globally over the week following its announcement.

Read the analysis →

1 story

Open issue →

Anthropic announces the Anthropic Public Benefit Fund

Anthropic announced the Anthropic Public Benefit Fund as a wholly owned limited liability company. Subject to stockholder approval, it intends to commit 5% of its shares to the fund. At its discretion, the board could add another 2% of shares for every 50% increase in company value above a set baseline. Anthropic says equity held by the entity will not be used for lobbying or political giving in the United States or abroad. It plans to provide further details about the fund’s governance and valuation milestones later.

Read the analysis →

3 stories

Open issue →

Google begins U.S. rollout of Universal Cart in Search and Gemini

Google has begun rolling out Universal Cart in the U.S. across Search and the Gemini app. Shoppers can save products from multiple retailers, receive alerts about price drops or items returning to stock, and see relevant promo codes. The cart can also help them choose a payment method using card benefits in Google Wallet. At supported merchants, shoppers can check out through the cart with Google Pay or move items to the retailer’s cart and finish the order on its site.

Read the analysis →

Google DeepMind introduces biomolecular structure viewing in Antigravity

Google DeepMind introduced Biomolecular Structure Viewer, a 3D tool embedded in Antigravity. A researcher and an agent can work with the same view of a structure: the agent can highlight regions, while the researcher’s annotations become available to the agent for follow-up discussion. In its published example, the tools were used to search for an antimicrobial peptide candidate. The candidate’s predicted activity is still undergoing laboratory validation.

Read the analysis →

Anthropic gives METR expanded access for AI risk assessment

Anthropic says it has onboarded six employees of the nonprofit METR with access to internal tools and systems roughly comparable to that of its own risk-assessment teams. They have also received company hardware and office space. METR may publish analyses of risk levels, incidents, and company practices. Anthropic cannot veto publication, though it can require the removal of information in narrow, predefined categories.

Read the analysis →

3 stories

Open issue →

Third Moment Research describes running 10,000 agents concurrently

Third Moment Research says it runs 10,000 agents concurrently on investigations lasting 36–48 hours. In its internal environment, the agents use OpenAI models and Codex tools, including to test trading strategies on historical data. The company also describes optimizing one strategy backtest: changes to caching and memory handling cut its runtime from approximately three hours to approximately three to five minutes.

Read the analysis →

Anthropic adds Claude safeguards against multi-step election abuse

Anthropic has announced new classifiers for detecting election-related misuse of Claude. They are intended to better identify multi-step schemes whose individual requests may look harmless. The company also says that, later in October, Claude.ai will begin incorporating Democracy Works voting information into responses for users in the United States. It has not described that response feature as already launched.

Read the analysis →

3 stories

Open issue →

Google begins global rollout of Gemini 4 Argon

Google announced a global rollout of Gemini 4 Argon for developers and enterprises through the Gemini API. Developers can use the API’s paid tier in AI Studio and Android Studio. Google AI Ultra subscribers can access the model in Antigravity; Ultra and Pro subscribers can use it in the Gemini app and AI Mode in Search. Access depends on the product and subscription: a global rollout does not mean identical terms for every user.

Read the analysis →

Google details Gemini 4 Argon safeguards and agent monitoring

Google described several layers of protection against dangerous uses of Argon: checks on requests and conversation context, analysis of the model’s internal signals, and monitoring while it produces a response. A separate monitor, the company says, examines an agent’s actions and step-by-step reasoning in real time to identify severe boundary violations. Google also began expanding its Fairwind Program to additional trusted cyber defenders; the program is not open to everyone.

Read the analysis →

Google Play starts enabling personal AI Studio app installation

Google began a phased rollout allowing people to install Android apps they created in Google AI Studio on their own devices. Installation goes through Google Play with a standard Google account; no developer account is needed. For now, the feature is for users aged 18 or over in selected markets, including the U.S. These apps remain personal and do not appear in Play’s public catalog.

Read the analysis →

3 stories

Open issue →

2 stories

Open issue →

Sophos reports results from OpenAI Daybreak agents in MDR

Sophos uses agents built through OpenAI Daybreak in its managed detection and response service, or MDR. One agent gathers customer context, detections, and threat information. A planning model creates an investigation plan, carries out its steps, and prepares a summary with recommendations for an analyst; other agents can perform parts of the response. Sophos reports an average response time of about 89 seconds for cases using agents, compared with roughly 38 minutes in its earlier process.

Read the analysis →

Asana releases changes to the StackAI browser agent

Asana released changes to browser navigation in StackAI: the agent retains more browsing history, caches that history, and removes screenshots in batches rather than after nearly every step. According to the published account, StackAI’s CTO directed experiments with GPT-6 Astra in Codex. In the tests, a GPT-6.1 Sol configuration completed the test task in about four minutes on average, at an estimated model cost of $0.47 per run.

Read the analysis →

3 stories

Open issue →

Anthropic introduces Cyber Mission for industrial systems and open-source code

Anthropic announced Cyber Mission as a long-term cybersecurity initiative. Its first two areas are the Critical Infrastructure Defense Program (CIDP) and OSS Scanner. CIDP brings together Claude models, on-site engineers, and threat research for providers trusted by operators to protect industrial equipment and networks. OSS Scanner offers enrolled open-source projects free periodic scans. Its reports describe how a suspected bug could be exploited and suggest a fix where one is available.

Read the analysis →

Google Cloud announces Gemini agent for work across business systems

At Gemini at Work 2026, Google Cloud introduced Gemini agent as a single interface for work requests. According to the company, it plans tasks, uses tools, and connects to Google Cloud customers’ business systems. It is meant to return results in familiar environments, including documents, inboxes, and developer tools. Google Cloud also claims model selection for each task, cost controls, and enterprise security and governance features. The announcement does not say who can access the agent or when.

Read the analysis →

OpenAI describes disrupting influence operations originating in Russia and Iran

OpenAI said it banned two influence operations that used its models alongside conventional ways of producing and distributing material. The Russia-origin operation, Dark Clark, involved a false-front research center in Latin America. The Iran-origin operation, Bogus Bylines, used seven fabricated author personas to pitch articles to online outlets and also prepared social-media comments. This is a report on misuse of models, not an announcement of a tool that automatically detects such operations.

Read the analysis →

3 stories

Open issue →

OpenAI starts rolling out GPT‑6 with Intelligent UI in ChatGPT

On October 7, OpenAI began a global rollout of GPT‑6 with Intelligent UI in ChatGPT’s Chat tab. Rather than responding only in text, the model can combine text, visuals, and elements people use within the conversation. The rollout starts with Plus, Pro, Business, and Enterprise; it is due to expand to Free and Go from October 8. Enterprise access depends on administrator settings. This release does not change the models powering Work or Codex.

Read the analysis →

Google opens global English-language access to SynthID Detector

On October 7, Google opened global access to SynthID Detector in English. The tool checks an image, video, or audio file for a SynthID watermark. Google says the check covers media made with its AI tools and those of participating partners, including OpenAI, NVIDIA, and Kakao. Apple is described as a future partner, not part of the stated current coverage.

Read the analysis →

Google launches Playground for creating games with text prompts

On October 7, Google launched Playground, an experimental browser-based platform for creating, playing, and publishing games. A user can describe an idea in a conversation, test the result, and ask to change rules, physics, characters, or the environment. The launch is for U.S. users aged 18 and over; creation access is rolling out in tiers based on Google AI subscriptions. A game can be kept private, shared by link, or published in a gallery.

Read the analysis →

3 stories

Open issue →

Atlassian and OpenAI expand their AI agent agreement

Under a new agreement, OpenAI models will power agents across Atlassian's platform and in Rovo. Rovo uses Teamwork Graph, an enterprise context layer connecting people, projects, documents, and decisions. Atlassian will also gain expanded access to OpenAI models, including GPT‑6 Astra and the GPT‑5.6 series. These are terms of an agreement, not an announcement that every proposed workflow is already live.

Read the analysis →

Google releases the multimodal EmbeddingGemma 2 model

EmbeddingGemma 2 represents text, code, images, video, and audio in a shared space of numerical vectors, or embeddings, that can be compared for semantic search. Built on Gemma 4, the model has 740 million parameters and is released under Apache 2.0; its weights are available on Hugging Face and Kaggle. Google proposes using it, among other things, for on-device search without sending source materials to the cloud.

Read the analysis →

Anthropic introduces three Cyber Verification Program access tiers

Anthropic announced an expanded Cyber Verification Program that brings it together with Project Glasswing. The program now has Defense Access for defensive work, Red Team Access for authorized penetration testing, and Specialized Access for a limited set of vetted organizations working with particularly sensitive systems. The tiers differ in applicant vetting, security controls, and the extent to which Claude's cyber safeguards are reduced.

Read the analysis →

3 stories

Open issue →

OpenAI opens optional text watermarking for API customers

OpenAI API customers worldwide can now opt in to textGrain watermarks for text from select models. The feature is off by default. In the coming weeks, the company plans to watermark eligible ChatGPT and Codex text output in the EU. Applications for detector access are open, but initial access is limited to approved researchers and expert organizations.

Read the analysis →

OpenAI introduces a visual ad format for ChatGPT

OpenAI introduced a visual ad format for ChatGPT. It plans to begin testing later in October with an initial group of US advertisers during image generation. Ads will be clearly labeled and separate from the image being created. OpenAI also expanded tools and partnerships for sharing conversion data, attribution and ad measurement.

Read the analysis →

Google expands AI training for UK and Ireland educators

Google expanded its free Google AI Educator Series to 650,000 educators in the UK and Ireland. Educators lead virtual and in-person events; participants can earn micro-credentials and badges for completing modules. Google also announced a partnership with the National Governance Association to create an AI Policy Toolkit for schools.

Read the analysis →

1 story

Open issue →

Anthropic launches Claude Frontier Academy for engineers

Anthropic has started the first Claude Frontier Academy cohorts in San Francisco, New York, and London. The program trains engineers to lead Claude projects within their organizations. It begins with an in-person simulation of an enterprise deployment and a practical assessment. Participants who pass move into a 12-week residency, working on a real project at their own organization with support from Anthropic engineers. A separate assessment follows that residency.

Read the analysis →

3 stories

Open issue →

Barclays expands Claude across banking operations

According to Anthropic, Barclays is expanding its use of Claude in software development, employee information retrieval, and client-request handling. In Global Markets, models help process incoming emails by classifying them, adding information about the request, and selecting a processing route. Barclays UK also has a live assistant that retrieves information for colleagues serving customers. Wider adoption of Claude Code among developers is an expected next step, not an achieved rollout.

Read the analysis →

Google launches Guided Vision in Gemini Live for Android

Google has launched Guided Vision for compatible Android devices. With the feature on, a person can share their camera during a Gemini Live conversation, hear descriptions, ask follow-up questions, and receive spoken prompts to adjust the camera for a clearer view. Google’s examples include reading fine print, finding objects, and describing nearby surroundings. Availability depends on more than the device: Gemini Live must also be supported in the user’s region and language.

Read the analysis →

Google’s Project Suncatcher prototype satellite reaches orbit

Google reports that a Project Suncatcher prototype satellite, built with Planet, reached orbit on SpaceX’s Transporter-18 mission. The company says it has established contact and the satellite is operating as expected. The next step is to gather data on how Google’s tensor processing units, or TPUs, used for machine-learning workloads, withstand the stresses of spaceflight. The launch announcement does not yet contain results from those tests.

Read the analysis →

3 stories

Open issue →

Google rolls out Gemini Skills and prepares to replace Gems

Google has begun rolling out Skills globally in Gemini chat. Users can save an instruction and invoke it by typing “/” followed by the Skill name. Skills can already include reference materials such as plain-text files, PDFs, and images, and users can combine several of them. Google also says Gemini can create a Skill from a chat and apply it to a matching request. Workspace Business, Enterprise, Nonprofit, and Education customers are due to receive access in the coming weeks; this does not mean every such customer has it today.

Read the analysis →

Google announces Gemini 4 Argon with limited initial access

Google has announced Gemini 4 Argon and begun providing access to a group of trusted cyber defenders through its Fairwind Program. It plans a wider release after early-user feedback and further work on safeguards, but gives no date. Google states that the model can produce up to 1 million output tokens, compared with a previous 64,000-token limit. Its announced introductory prices are $2 per million input tokens and $10 per million output tokens; cached input tokens are priced 95% below regular input tokens. Google Blog lists post-introductory prices of $4 and $20 respectively.

Read the analysis →

OpenAI describes attempts to extract hidden model reasoning and its defenses

OpenAI says it disrupted coordinated attempts to obtain hidden model reasoning for distillation: using one model’s output to help train another. Its disclosure concerns past activity, not a new product. The company says it closed paths for repeatedly obtaining reasoning, strengthened account and infrastructure controls, and inspected streamed output. OpenAI also shared findings with partners but says defenses in partner deployments and for data from connected tools still need work.

Read the analysis →

3 stories

Open issue →

OpenAI releases GPT-6.1 Sol with API access

OpenAI introduced GPT-6.1 Sol as an upgrade to GPT-6 Sol. It is available in the API as gpt-6.1-sol and to Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex. It is not yet in Chat. Standard API prices are $2 per million input tokens and $10 per million output tokens; cached input costs $0.10 per million tokens. OpenAI plans to add an Ultrafast tier in the coming days.

Read the analysis →

OpenAI adds computer use to the Agents API

In its DevDay recap, OpenAI said the Agents API now supports computer use, allowing agents to operate software through its interfaces. It also added capabilities from Codex: multi-agent coordination, tool search and calling, and context compaction. OpenAI runs the underlying infrastructure. A separate Decisions API is designed to select an answer from a finite set of predefined options; it has a different availability status.

Read the analysis →

OpenAI begins a limited rollout of persistent dots assistants

OpenAI introduced dots, persistent assistants powered by GPT-6 Astra. Each dot has its own cloud computer and browser. OpenAI says a dot can work across projects, learn from feedback, and use connections from an ecosystem of more than 4,000 apps. Users can contact dots in ChatGPT, Slack, or Teams, inspect their work, set boundaries and app permissions, and review actions that need approval. The rollout covers eligible markets for ChatGPT Pro and Business Premium users.

Read the analysis →

3 stories

Open issue →

OpenAI releases GPT-6 Sol and Luna and updates prompt caching

OpenAI introduced two GPT-6 models. In the API, Sol costs $2 per million input tokens and $10 per million output tokens; Luna costs $0.10 and $0.50 respectively. The caching update adds a dashboard for monitoring reuse, diagnostics for cache misses, and explicit breakpoints for choosing which prompt prefixes to reuse. OpenAI also describes ways to change reasoning effort and tool availability without disrupting reuse of eligible context.

Read the analysis →

Anthropic releases Claude Opus 5.5 with new prices and safeguards

Claude Opus 5.5 is available on Anthropic’s platforms and through AWS, Google Cloud, and Microsoft Azure. Standard pricing is $4 per million input tokens and $20 per million output tokens; cache reads cost $0.20 per million. A separate fast mode is available in Claude Code and the Claude Platform at $8 and $40 respectively. The release includes safeguards for cybersecurity and biology tasks; for API accounts created on or after August 31, 2026, preserved thinking prevents edits to prior model context intended to extract its reasoning.

Read the analysis →

3 stories

Open issue →

Anthropic and Accenture announce independent AI evaluation from inside the lab

Anthropic announced a non-exclusive partnership with Accenture for independent evaluation of frontier AI models. Faculty, Accenture’s specialist AI business, will lead the work. Plans include model evaluation, red-teaming, alignment assessments, and safeguard testing. According to Anthropic, each company expects to invest at least $1 billion in building capacity in this area over the next five years; these are expectations, not a report of money already invested.

Read the analysis →

Google describes two Flow tools for fashion-show preparation

Google says its Envisioning Studio, supported by Google Labs, co-developed two specialised Flow tools with Jane Wade and Sergio Hudson ahead of New York Fashion Week. According to Google, Styling Suite helped Wade assemble digital looks before making additional garments. The company says Runway Visualization allowed Hudson to simulate the venue, change lighting and props using options within his budget, and refine the models’ routes.

Read the analysis →

OpenAI publishes an Australian roadmap for teen safety

OpenAI published the Australian Youth Safety Blueprint, a roadmap for protecting young AI users. The company describes six pillars, but the supplied text names only five: AI literacy, age-appropriate safeguards, privacy-protective age assurance, connections to real-world crisis support, and accessible parental controls. OpenAI also said it began rolling out ChatGPT for Teens in Australia in August, as the default experience for users identified as aged 13–17.

Read the analysis →