The Hal Times

Anthropic is operating a lab that conducts biology experiments, TechCrunch reports. The work sits beside the company’s better-known AI research, though the report does not say where the lab is, how many people it employs, or which experiments it conducts. AI leaders have been promising that AI is the key to curing human disease. Anthropic researchers have also been warning that AI might kill us all. The combination gives the account an unusual cast. It joins practical biological work to the same company’s warnings about the risks of advanced systems.

That is the part actually new in the account: Anthropic is not only discussing biology as a possible use of AI, but conducting biology experiments in a lab of its own. The material supplies no research result, benchmark, model name, or timetable. It therefore does not show whether the work has produced a medical advance, a useful research tool, or merely an early experimental programme. Nor does it establish how closely the lab’s work is tied to Anthropic’s models. Those omissions matter more than the grander promises surrounding the field.

For the moment, the lab is a fact about Anthropic’s operations, not a result for medicine. The company’s public posture contains a plain tension: AI is presented as an instrument for curing disease, while AI safety concerns include the possibility that it could kill us all. The report leaves the reader without a method for weighing those claims against the new work. Until Anthropic discloses its experiments and findings, the biology lab tells us where the company is looking, but not what it has found. (TechCrunch)

Vantora Raises $100M For Physical AI

UP.Labs, now doing business under the name Vantora, has raised $100M for a business devoted to physical AI, TechCrunch reports. The company builds startups for industrial corporations rather than presenting itself as a single application vendor. That arrangement makes the customer, not merely the model, part of the enterprise from the outset. The report gives no account of the investors, the corporations involved, or the startups already formed. It does, however, supply a useful distinction: Vantora is selling the making of companies as its business.

The sum is substantial, but the account offers no valuation, revenue figure, product list, or evidence of commercial results. Nor does it explain what Vantora means by physical AI in practice, beyond its work with industrial corporations. The company’s plan is therefore clearer than its record. It has chosen a structure for producing ventures and found $100M to support it. Whether that model can repeatedly turn industrial problems into durable companies remains unknown, as does the number of businesses Vantora intends to build. (TechCrunch)

Anthropic Targets November IPO At Potential $2 Trillion Valuation

Anthropic targets a November IPO at a potential $2 trillion valuation, according to Crypto News. The report gives the market a large number and a near date, but little else. It does not establish whether Anthropic has filed, selected advisers, fixed a price, or confirmed the timetable. Nor does it say how the potential valuation was reached. For now, the figure belongs to a report about the company’s intentions, not to a completed transaction. The distinction is modest in wording and important in finance, where November remains a month rather than a guarantee. (Crypto News)

From the Laboratories

•  Bloomberg.com reports that a China State TV Affiliate flags Anthropic data and privacy risks. (Bloomberg.com)

•  OpenTheMagazine reports that Anthropic, OpenAI, and Google face a US antitrust lawsuit over questions of AI safety or collusion.

•  Storyboard18 reports that Anthropic weighs a new model to counter OpenAI’s Astra, days after its CEO called for an AI slowdown. (Storyboard18)

•  OpenAI introduces the Australian Youth Safety Blueprint, a six-pillar roadmap for safer AI experiences that protect and empower young people. (OpenAI)

Antitrust Legal Action

Firms Face Federal Antitrust Lawsuit Over AI Slowdown

Anthropic, OpenAI, SpaceXAI, and Google are now the subjects of a federal antitrust lawsuit. The plaintiffs allege that the companies, by publicly advocating for a slowdown in artificial intelligence development, have engaged in anti-competitive behaviour. These major firms have frequently called for increased regulation and industry-wide pauses to address safety concerns, but legal observers note that such coordination can invite scrutiny under competition laws. The central question is whether these safety declarations serve the public interest or act as a mechanism to suppress competition by raising barriers to entry for smaller, independent rivals.

The lawsuit suggests that by working together to set the pace of innovation, these companies may be limiting the options available to the market. This claim challenges the prevailing narrative that the firms are solely motivated by ethics and risk management. As these organisations have become the primary architects of the industry, their ability to dictate the tempo of progress effectively allows them to manage the landscape to their own advantage. Whether this conduct amounts to illegal collusion or legitimate self-regulation remains for the court to decide, though the outcome will dictate how developers operate in the future.

The legal community remains divided on the strength of these arguments. Some suggest that public statements alone rarely meet the threshold for antitrust violations unless they result in binding agreements that harm consumers. Conversely, others argue that when the largest players in a nascent field act in unison, they create a chilling effect on innovation that demands regulatory correction. With the suit moving forward, the industry waits to see if the courts will impose limitations on how these companies discuss the security of their technology. (The Tribune; PGurus)

California Governor Issues Executive Order On AI

Governor Gavin Newsom has issued an executive order focused on the trajectory of artificial intelligence within California. The Electronic Frontier Foundation has welcomed the move, noting that it provides a formal venue for discussing potential harms and public anxieties. The organisation suggests this initiative is a necessary step for the state to manage the rapid deployment of these technologies while safeguarding civil liberties. By creating a regulatory framework, the state intends to lead the national conversation on responsible development.

The order serves to formalise the state's engagement with industry stakeholders and safety advocates. While the policy signals a shift toward oversight, critics note that translating such orders into effective governance remains a challenge. The long-term impact on how companies develop and release tools in California is yet to be determined, as the specifics of the directive are currently being interpreted by agencies. (EFF)

Parole Decisions Subject To Review

Australian authorities are reviewing the use of artificial intelligence in parole decision-making. This move follows recent revelations concerning the case of Neill-Fraser. Officials intend to assess whether current automated systems align with established legal standards and provide necessary transparency for those impacted. The review aims to ensure that algorithms do not inadvertently compromise the fairness of judicial outcomes or the rights of prisoners.

The investigation highlights the growing tension between the efficiency offered by automated tools and the human oversight required in the justice system. The future of these tools in Australian courts remains uncertain until the inquiry concludes. (ABC News & Headlines)

In Brief

•  OpenAI faces scrutiny for alleged discrepancies between its public calls for regulation and its actual adherence to existing legal statutes. (Dagens.com)

•  Massachusetts officials express a commitment to balancing AI innovation with the protection of resident rights. (MassLive.com)

•  Georgia's attorney general has joined the ongoing policy debate regarding the appropriate scope of AI regulation. (Yahoo)

•  The Electronic Frontier Foundation advises lawmakers to root AI cybersecurity legislation in proven best practices following recent lab security incidents. (EFF)

Tools And Agents

DoorDash Deploys Multi Agent System For Cleanup

The engineering department at DoorDash has successfully implemented a multi-agent system to address technical debt. The firm utilised this automated workflow to prune over 60,000 stale feature flags distributed across 623 repositories. By combining live experimentation data through MCP with isolated Git worktrees and parallel agents, the team managed the cleanup process with human oversight reserved for final approvals.

During an initial evaluation involving 50 flags, the system generated 45 usable pull requests. The process required an average time of 13.8 minutes per cleanup at a financial cost of $4.79. This suggests a repeatable method for maintaining large-scale codebases.

The firm integrated engineer approval cycles to ensure the automated output met necessary quality standards before merging. The project highlights a shift toward using agents for routine maintenance tasks that were previously performed manually by staff. It remains to be seen how well such systems scale as the number of repositories and agents continues to increase within large corporate environments. (InfoQ)

WSO2 Releases Agent Manager

WSO2 has announced the general availability of WSO2 Agent Manager, an open-source platform intended to address the rise of AI agent deployments in enterprise settings. The tool provides a centralized hub for governance, identity management, and security controls.

As businesses deploy agents across varied models and frameworks, oversight has become a technical challenge. This platform offers operational visibility for agents regardless of their underlying infrastructure. It aims to provide the necessary structure to manage agent sprawl while maintaining security standards within corporate networks. (InfoQ)

Claude Code Adds AGENTS.md Support

The latest update to Claude Code, version 2.1.277, introduces support for AGENTS.md files. If a project folder lacks a CLAUDE.md file, the tool will now automatically check for and utilise the instructions found in AGENTS.md.

This feature is built upon the Claude Code mods harness, which allows developers to customise project instructions. The company notes that this is a built-in mod, and users will be able to construct their own project-specific instructions to refine how the agent behaves within a given directory. (Simon Willison)

In Brief

•  Reinventing.AI has released eight open source AI employees on GitHub under the MIT license. (EIN Presswire)

•  IBM and NASA have released an open-source AI model intended to support lunar exploration. (Portal ERP)

•  Alibaba has open-sourced a medical AI model capable of detecting cancer and nearly 150 other conditions. (SCMP)

•  OpenAI has introduced a new disclosure framework to report model misalignment and unexpected behaviours. (InfoQ)

Content Under Scrutiny

OpenAI And Microsoft Face Fresh Legal Scrutiny Over Data Practices

Newly unsealed court documents from the ongoing lawsuit brought by the New York Times against OpenAI and Microsoft reveal internal concerns regarding the development of artificial intelligence systems. Company documentation suggests that executives were aware of the risks posed to the web by their training processes, describing the potential outcome as a doom loop. The documents also contain internal characterisations of the practice of scraping data for model training as the largest theft of labour in human history.

The revelations come as the industry faces increasing pressure from publishers and creators who argue that their intellectual property has been utilised without proper authorisation. Executives at both companies reportedly grappled with the question of whether their systems were built on stolen content. The case highlights the tension between the rapid expansion of generative models and the established rights of those who produce original creative work.

These court findings provide a glimpse into the internal deliberation of firms currently at the forefront of the technology. While the companies continue to deploy their models, the legal implications of their early data gathering strategies remain under review. The extent to which these internal warnings will influence the judicial outcome of the case is not yet known. (The Verge; Hollywood Reporter)

Japanese Game Developers Adopt Generative AI

A recent report from the Tokyo Game Show indicates that generative AI has seen widespread adoption within the Japanese gaming industry. Almost 86 percent of developers in the region now incorporate these tools into their workflows. The figures highlight a significant shift in how titles are designed and produced in a market known for its high standards of craftsmanship.

Despite this trend, the use of AI remains a subject of public scrutiny. Insomniac Games recently issued a confirmation that its title Wolverine does not contain AI-generated assets, following speculation sparked by a sign spotted in promotional material. The studio moved to address concerns from fans who remain wary of automated content in big-budget games. (TweakTown; Kotaku)

Disney Appoints Chief Technology Officer

The Walt Disney Company has appointed Karandeep Anand as its first-ever chief technology officer. Mr. Anand, who previously held positions at Meta and Microsoft, will oversee the studio as it adapts to the generative AI era. The move signals a formal commitment to integrating new technologies into the company's global entertainment operations.

This appointment arrives as Hollywood labour groups continue to express caution regarding the use of AI. Organisations have urged the public to focus on the immediate effects of these tools on current creative production, rather than solely on existential warnings presented by the tech sector. (Hollywood Reporter; The Verge)

In Brief

•  A new 2.5-hour motion picture generated entirely by artificial intelligence has received negative critical reception. (gnnhd.tv)