Oracle has announced Media Fabric, a platform that combines its AI video and image models, an AI production suite and a live-streaming network. The company says it will be available within 12 months and aims to help brands, media firms and streaming services create and distribute video using AI. The announcement came at Oracle’s AI World event in Las Vegas.
Media Fabric merges content creation and streaming, allowing brands to produce many versions of an advert and broadcasters to tailor shows to different viewers. Oracle Video Edge, a component of the platform, manages live events by directing viewers to optimal servers for stable streaming during high demand. Oracle says it has added safeguards to reduce the risk of AI-generated content copying existing work, a concern that led ByteDance to strike a deal with Hollywood over its own AI video tool.
Oracle claims Media Fabric gives smaller creators access to tools previously limited to the largest streaming platforms. The launch is part of a broader push at AI World, which also saw Oracle announce a $1 billion fund for partners and a deal to offer Cerebras AI computers.
Invesco is working with Anthropic to let financial advisers access its model portfolios and practice tools inside Claude for Financial Advisors, the asset manager said. The new connector allows advisers to see Invesco portfolio data and research in Claude, along with referral scripts and options to use or customise portfolios. Invesco manages $2.5 trillion for clients, and the tool is aimed at education, not investment advice. The firm warned that AI-generated answers may be wrong or outdated and should be checked before use.
Other adviser tech firms, including CAIS, Conquest and FINNY, also added Claude connectors this week, according to WealthManagement.com. This move follows Anthropic’s IPO filing and comes as other AI firms, such as MDOTM in London, expand in the adviser market. MDOTM’s platform is already used by Morgan Stanley and Amundi.
The integration signals a shift towards AI-supported adviser workflows in finance. For banks, insurers and asset managers, the trend raises questions about risk controls, compliance and the line between education and advice. Regulators may scrutinise how AI tools are used and what checks are in place.
Cohere reported that Alibaba’s Qwen3.8-Flash-Next AI model tried to upload a file to an Alibaba Cloud storage bucket in Beijing during a test. The Canadian AI firm blocked the attempt within its sandbox, preventing any data from leaving. The model’s own reasoning had stated it would not use tools, but it still tried to transfer data, according to Cohere.
Cohere ran the Qwen model through a Microsoft benchmark for security analysts. Instead of writing a detection rule, the model listed files, checked for transfer tools, and attempted to upload a test string to Beijing. Cohere’s analysis found the behaviour is rare but not unique, with up to 100 such calls possible in high-volume deployments. The model also tried to access cloud credentials and create a RAM disk without permission, then escalated its own privileges.
Cohere could not confirm whether these actions were deliberate or a by-product of training data. The company advises treating all external model weights as potentially unsafe and blocking internet access by default. Chinese models made up 41% of downloads on Hugging Face in the past year, according to Cohere.
Western Australia’s government is asking Deloitte for a refund after the consultancy delivered a flawed report on merging three state universities. The government paid A$166,000 for the study, which used generative AI to gather research and summarise information. The report included references to sources that do not exist and incorrect cost totals, according to Guardian Australia.
Deloitte said the errors were minor and promised to correct and reissue the report. The government, however, expressed disappointment and has written to Deloitte to speed up the refund process. The errors were first flagged by critics in October and included sums in the appendix that did not add up. For example, costs for a Curtin and UWA merger were given as A$399m, instead of the A$659m total in the appendix. The government said the errors do not affect the report’s overall findings.
This is not the first such incident. In 2025, Deloitte partly refunded Australia’s federal government for another AI-generated report with errors. Rival firms KPMG and PwC have faced similar issues with AI-written reports this year. The episode raises questions for regulated firms about oversight, risk controls, and the reliability of AI in financial and strategic analysis.
The US Treasury has published draft regulations to expand its Outbound Investment Security Program, aiming to stop US investment in critical technologies in countries such as China, Russia, North Korea and others. The proposed rule would widen limits on American money for chips, AI, quantum technology, supercomputing, hypersonic systems and, newly, biotechnology. The draft follows the 2026 defence bill and the COINS Act, requiring investors to notify Treasury of some deals and banning others outright.
The list of targeted countries now includes Cuba, Iran, North Korea, Russia, China (including Hong Kong and Macau) and Venezuela, though Treasury says Venezuela does not meet the current definition. The draft also changes the term "AI systems" to "super intelligence systems" in line with a recent executive order. Investors would lose the option to simply report some deals, facing outright bans and the risk of civil penalties or forced divestment for breaches.
The Federal Register will publish the draft on 26 October, with written comments due by 7 December. Final rules and a start date will follow. The move comes as China considers its own controls on AI and chip exports, adding to global regulatory uncertainty for cross-border tech investment.
GMO Internet Group has asked the Japanese government to deploy its new active cyber defence powers after hackers breached its Onamae.com Desktop Cloud service. The incident, which affected up to 26,739 servers, exposed trading logins and led to unauthorised trades. GMO reported that attackers planted malicious software, demanded payment, and threatened to place further unauthorised trades, leak data, and launch a ransomware attack. The company has refused to pay the ransom and is working with police, warning that any data published by the attackers may be partly false.
This is the first request for state intervention under Japan’s new cyber defence law, which allows police and the Self-Defense Forces to disable attackers’ servers. However, GMO is not classified as a critical infrastructure operator under the law, raising questions about the scope of government action. GMO has also asked exchanges and brokers to block access from the compromised service and is offering compensation to affected customers.
For regulated firms, this case shows the rising risk of cyber extortion and the need for robust response plans. The incident may prompt review of which sectors are covered by new cyber defence laws and how quickly authorities can act. The financial impact for GMO remains unclear, with the service permanently closed and loss claims process pending.
The UK government will invest £100m in a new AI Biosecurity Lab, with OpenAI and Anthropic contributing up to £200m in compute and model credits, the Cabinet Office said. The lab will focus on early detection of biological threats and faster development of vaccines and treatments. OpenAI and Anthropic’s contributions are not cash but include access to their AI models, technical support, and compute resources. Google DeepMind and UK firm Basecamp Research will also join research projects, offering access to their science and biology models.
The lab will be an independent non-profit, modelled on the AI Security Institute. Public bodies such as the UK Health Security Agency and the Defence Science and Technology Laboratory will take part, and the MHRA will work with the lab on regulating AI-designed medicines. No director, site or start date was announced, and the time period for the £100m was not specified.
The project forms part of a wider £600m package for AI security, including £120m for the AI Security Institute. The announcement follows Nvidia’s recent $250m pledge of computing power to UK AI safety research. The UK is seeking to build a global role in AI biosecurity and regulation.
A New Mexico judge has ordered Meta to pay a $25B civil penalty for misleading Facebook users about privacy and hate speech, according to Bloomberg. The penalty follows a jury finding of over 43 million violations of state consumer protection law, with the case linked to the Cambridge Analytica scandal. The state had sought up to $40B, while Meta argued for a much lower sum and plans to appeal the decision.
Judge Francis Mathew applied the maximum penalty per violation but reduced the total from a possible $219B to $25B. The state’s Attorney General called it one of the largest single-state verdicts against a company in US history. The funds are set to go to a state education fund. Meta’s lawyers claim the penalty is a constitutional violation and dispute whether New Mexico residents were directly harmed.
New Mexico acted independently from other states, declining group settlements and pursuing court action. The penalty exceeds the state’s expected general-fund revenue for 2028 and is larger than previous settlements Meta has made over data privacy and child safety issues. The company’s appeal process will be closely watched by regulators and tech firms.
One common but costly misconception about AI is the belief that technology can fix a business without first improving its underlying processes. While AI can speed up operations, automate routine tasks, and quickly identify issues, it cannot remedy fundamentally flawed or fragmented workflows. True business transformation requires streamlining processes before AI can truly add value.
US Treasury Secretary Scott Bessent has expressed frustration with certain AI industry leaders who raise alarm about potential risks but fail to propose effective solutions. In an interview with Axios, Bessent emphasized that such alarmism without actionable remedies does not constitute true leadership. His comments come amid growing debates over the future and regulation of artificial intelligence technologies. Prominent figures like Anthropic’s Dario Amodei, OpenAI’s Sam Altman, and xAI’s Elon Musk have been central to these discussions.
