The Hal Times

Anthropic announced that it has disrupted an operation that used artificial‑intelligence tools to manipulate voters in Malaysia. The firm said the campaign targeted social‑media users with tailored political content generated by its models. It identified the actors through unusual query patterns and halted the accounts before further influence could be exerted. The disruption was reported to Malaysian authorities, and Anthropic shared technical details with industry partners to improve detection. The claim marks the first public acknowledgement of an AI‑driven election interference effort.

The episode underscores growing concerns that large language models can be weaponised for political ends. By intervening early, Anthropic demonstrates a proactive stance that may encourage other providers to embed monitoring tools in their APIs. Regulators are watching the incident closely, and the Malaysian election commission has pledged to review its digital‑campaign policies. However, the episode also reveals the difficulty of tracing AI‑generated influence when actors operate across borders and hide behind legitimate cloud services.

What remains unclear is the scale of the operation and whether similar campaigns are already active elsewhere. Anthropic did not disclose how many queries were flagged or the exact methods used to identify the actors. Analysts question whether the disruption was a single incident or part of a broader, coordinated network exploiting AI for disinformation. Further transparency from the company and cooperation with international election monitors will be required to assess the true impact on democratic processes. (CNA)

Anthropic Details Distillation Campaigns From Alibaba, Moonshot AI and DeepSeek

Anthropic released a report describing ongoing distillation attacks launched by three China‑based AI firms – Alibaba, Moonshot AI and DeepSeek. The company says the attackers repeatedly query its models to extract proprietary weights, then fine‑tune the distilled versions for commercial use. The report cites an increase in such attempts over the past months, with each firm mounting dozens of campaigns that evade standard rate‑limit safeguards. Anthropic alleges that the attacks compromise intellectual property and could accelerate the spread of powerful models without oversight.

Anthropic has responded by tightening its API monitoring and introducing a watermarking scheme to detect illicitly derived models. It is also collaborating with industry peers to develop shared signatures that identify distilled copies. The effectiveness of these measures remains to be proven, as the report notes that attackers adapt quickly to new defenses. Observers note that the episode highlights a nascent arms race in model protection, where commercial incentives may clash with open‑source ambitions. (TechCrunch)

OpenAI Launches Data Agent for ChatGPT Work

OpenAI unveiled a new Data agent within ChatGPT Work, allowing users to connect corporate data sources and query them in natural language. The feature creates interactive dashboards and surfaces insights without requiring code, promising to accelerate business intelligence workflows. OpenAI markets the agent as a secure bridge between internal repositories and the language model, with access controls that restrict data exposure. The launch follows a series of enterprise‑focused updates aimed at expanding the platform’s utility beyond research labs.

The agent’s performance will depend on the quality and structure of the connected datasets, and OpenAI has not disclosed latency or cost metrics for large‑scale deployments. Security analysts caution that exposing internal tables to a language model could introduce new attack surfaces, especially if the model is later fine‑tuned on the retrieved information. OpenAI promises ongoing audits and an opt‑out mechanism, but the real‑world robustness of the Data agent remains to be demonstrated. (OpenAI)

From the Laboratories

•  Anthropic says it stopped AI misuse for cyberattacks and bioweapons. (Euronews.com)

•  Anthropic and OpenAI employees voice concerns about humanity’s extinction as debate intensifies. (Business Insider Africa)

•  Jensen Huang says Nvidia expects 70% growth next year. (TechCrunch)

•  Mark Wahlberg will appear at TechCrunch Disrupt 2026 to discuss entrepreneurship. (TechCrunch)

•  OpenAI pauses new Pro subscriptions as Astra demand strains its systems. (TechCrunch)

•  Researcher uses Codex and ChatGPT to hunt for antimicrobial molecules in genomes. (OpenAI)

•  OpenAI introduces ChatGPT for Financial Services, powered by GPT‑6 Astra. (OpenAI)

•  OpenAI and GSA offer free licences and discounts to US government entities, expanding AI access and cyber‑defence support. (OpenAI)

•  MIT Technology Review reports that a transmission fault in Ashburn, Virginia removed over 3 GW of load, highlighting AI’s power‑grid challenges. (MIT Tech Review)

•  Hugging Face publishes a guide on rebuilding AUTOMATIC1111 with Gradio Workflow. (Hugging Face)

Calls For Oversight

Urgent Demands For Artificial Intelligence Regulation

Legislators in the United States and abroad are moving with renewed focus to establish a framework for the governance of artificial intelligence. Recent discourse reflects a deepening anxiety regarding the trajectory of the industry, with some groups citing existential risks as a primary justification for immediate government intervention. The debate now spans the mechanics of cloud infrastructure, the concentration of power in large-scale computing, and the fundamental safety of deployed models.

While some officials advocate for strict control to ensure public safety, the legislative path remains fractured. Senator Ted Cruz has recently drawn notice for his public statements on the subject, highlighting the contentious nature of the debate in Washington. Across the Atlantic, European officials continue to centre their strategy on the concept of digital sovereignty, seeking to leverage regulation to maintain control over data, semiconductors, and platform infrastructure.

The rapid growth of the sector poses a significant challenge to existing regulatory bodies. As companies push for faster development cycles, observers note a tension between industrial momentum and the cautious pace favoured by those who fear systemic harm. Whether comprehensive policy can keep step with the capabilities of modern large language models remains an open question for the coming session. (Devdiscourse; Yahoo News Singapore; EFF; KuCoin)

Investors Target Datavault AI Inc In Securities Suit

Investors are seeking to lead a securities fraud lawsuit against Datavault AI Inc. The action follows allegations of misconduct that have prompted legal scrutiny into the company's financial disclosures. The litigation aims to recover losses for shareholders who claim they were misled by the firm's representations of its technological capabilities.

The court case arrives as regulators continue to examine the transparency of artificial intelligence ventures. The outcome may set a precedent for how firms are held accountable for their public statements in a rapidly evolving market. (Morningstar)

In Brief

•  OpenAI is pursuing a measured approach to the pace of development, a strategy intended to balance industry growth with safety considerations. (TradingView)

•  A series of power grid failures in Virginia serves as a reminder of the infrastructure challenges facing the artificial intelligence sector. (MIT Tech Review)

•  Public demonstrations involving humanoid robots have been staged to draw attention to the demand for formal artificial intelligence regulation. (News.com.au)

Model Releases

OpenAI Releases GPT-6 Astra For Coding And Computer Use

OpenAI has introduced GPT-6 Astra, a model designed for complex software development and machine interaction. The system is optimised for coding, computer use, and long-running agentic tasks. These capabilities are intended to assist developers in managing cybersecurity workflows and automated programming challenges. The model is currently available through the ChatGPT interface, the Codex platform, and the OpenAI API.

The release arrives as development teams continue to transition from simple code generation to more verification-heavy workflows. By focusing on agentic behaviour, the model aims to handle tasks that require persistent focus on a codebase. It remains to be seen how the model performs in large, legacy environments compared to human-managed specifications.

As AI coding assistants become standard in daily development, the industry is increasingly focused on the security implications of such tools. With GPT-6 Astra, OpenAI seeks to provide a balance between the speed of automated generation and the rigour required for secure, functional software. Future updates will likely clarify how this model interacts with existing security guardrails and external verification tools. (InfoQ)

Cognition Reports New Benchmark Results For SWE-2

Cognition has announced that its latest model, SWE-2, has achieved a score of 92.8 on the Terminal-Bench 2.1 evaluation. This result marks a significant development in the capabilities of agents designed for autonomous software engineering.

Terminal-Bench tests an agent's ability to navigate terminal environments and perform multi-step programming tasks without manual intervention. The high score suggests that SWE-2 is increasingly capable of managing standard developer workflows through command-line interfaces. Further scrutiny will be required to determine the model's reliability in handling novel, edge-case debugging scenarios that fall outside of current benchmarks. (Tokenstead)

IBM And NASA Release Open-Source AI Model

IBM and NASA have released a new open-source AI model intended to assist with lunar exploration. The model is designed to process data sets relevant to the lunar environment, potentially aiding researchers in mission planning and analysis.

The collaboration emphasises the role of open weights in scientific research. By providing the model for public use, both organisations aim to accelerate discovery in aerospace and planetary science. The model is available for inspection and deployment by the broader developer community, though its specific utility in extreme operational conditions remains a subject of ongoing testing. (HPCwire)

In Brief

•  System76 has introduced the Thelio Mira AI Linux workstation, which features up to 192 GB of GPU memory for intensive local AI tasks. (System76)

•  The GitHub Copilot app now supports side-by-side viewing of diffs, terminal commands, and web previews to reduce tab switching. (GitHub)

•  Google has launched a new Google Cloud developer plugin to integrate its AI coding assistant directly into cloud-native workflows. (Bitget)

Music Licensing Shift

Universal Music Partners With ElevenLabs

Universal Music Group announced a multiyear licensing agreement with ElevenLabs on Thursday. The partnership aims to build an AI-powered platform that permits users to create remixes, mashups, and new versions of tracks from the label's extensive catalog. This move marks a significant shift in how major record labels manage their digital assets in the age of generative models.

The development follows a broader trend of industry players seeking to control the use of their intellectual property within AI environments. By partnering with ElevenLabs, which specialises in audio generation, Universal intends to offer sanctioned creative tools to fans and creators.

The technical specifics of how the platform will safeguard original artist rights remain unclear. Similarly, the impact this will have on traditional royalty models for remixes remains an open question for the industry. The platform represents an attempt to formalise a practice that has previously existed largely in the grey areas of internet content creation. (The Verge)

AI Tools Disrupt Hollywood Microdramas

The market for phone-centric soap operas in Los Angeles faces significant contraction as AI-generated content floods mobile applications. These microdramas previously provided essential work for cast and crew following recent labour strikes in the industry.

As automated content production becomes more prevalent, opportunities for human actors and production staff are disappearing. The rise of these AI-generated verticals suggests that the niche medium, which served as a vital safety net for many, may no longer be viable for human creators seeking professional employment. (Hollywood Reporter)

Director Danny Boyle Employs Generative AI

Director Danny Boyle has confirmed his use of generative AI in the production of his latest film, Ink. The director utilised the technology specifically to animate static photographs during the editing process.

This public acknowledgement stands out in an industry currently navigating the ethical and creative boundaries of machine-generated imagery. While Boyle's use was confined to photo animation, the disclosure highlights how AI tools are being integrated into professional cinematic workflows in ways that were previously unavailable to filmmakers. (PetaPixel)

In Brief

•  A controversial film featuring AI technology has made its debut at the Venice Film Festival. (cphpost.dk)

•  A viral video depicting an Indian cricketer performing a complex catch has been confirmed as AI-generated. (LatestLY)

•  The inaugural AI Music Awards will take place in Los Angeles this November. (Lelezard)

•  Ticketmaster has been named the first music industry partner for Meta’s new Muse AI agent. (Music Ally)