Eleven AI stories from Monday daytime, condensed with sources linked. This is a news roundup, not original reporting.

1. Meta plans to raise $95 billion from shareholders to pay for its AI build-out

Meta said on Monday it will raise a total of $95 billion for AI infrastructure and what it calls "personal superintelligence". It is starting with about $35 billion of underwritten public offerings, roughly $22.5 billion of ordinary shares and $12.5 billion of a new mandatory convertible preferred stock, followed by a planned programme to sell up to $60 billion more shares on the market from early 2027. Meta said Nvidia, AMD and Broadcom were interested in buying up to $5 billion of the shares. Last week Meta raised its 2026 spending forecast to $135–145 billion and reported negative free cash flow; its shares rose 1.6% before the market opened.

Sources: CNBC, Meta Investor Relations

2. Security start-up Cathedral says its AI agents found a way into Anthropic's internal systems in under 12 hours

The Washington Post reviewed a draft report by Cathedral, a months-old cybersecurity firm founded by former US DOGE staff and valued at $1.4 billion. Testing Anthropic under its bug-bounty programme, Cathedral ran AI agents for less than 12 hours and found routes through flaws in third-party business software to internal data such as meeting transcripts, salary and job-offer details, and Slack messages about Anthropic's models. Anthropic confirmed access to "a limited set" of sales, HR and IT systems, said customers' chats, its code and its model weights were not reachable, and said it shut off the access the same day, disabled the software until vendors fixed it, and briefed federal officials. It also said Cathedral overstated the national-security angle and did not fully follow the programme's rules. A senior Trump administration official said the report had raised concern inside the government.

Sources: Washington Post, DNYUZ

3. US election campaigns are now writing content to make ChatGPT, Claude and Gemini recommend their candidate

A Reuters exclusive, based on documents it saw, describes "generative engine optimisation" (GEO), the AI-era version of search-engine tricks. A pitch deck prepared for Mike Minogue, the Republican challenging Massachusetts Governor Maura Healey, proposed creating hundreds of AI-friendly policy pages, paying for placements on websites that chatbots often cite, posting on Reddit and YouTube, and tracking chatbot answers and changing content until they reflect the candidate's platform. Fees range from hundreds of dollars a month to tens of thousands. About 25% of eligible US voters have asked a chatbot about a candidate, according to the National Conference on Citizenship. The Minogue campaign did not respond.

Source: Reuters via Investing.com

4. SpaceX signed the $40 billion AI-chip deal with Apollo — follow-up to the morning story

New development on this morning's story about Musk's $40 billion chip financing. In a filing with the US Securities and Exchange Commission on Monday, SpaceX said that on 24 Oct it signed a binding commitment for an entity managed by investment firm Apollo Global Management to buy GPU pods and related data-centre equipment and lease them to SpaceX. Each batch is leased for about five years, with rent paid quarterly, and total principal payments are expected to be about $40 billion. The deal is expected to close before the end of 2026, with chips delivered and leases starting in the first half of 2027. It confirms who is lending and how: the chips sit with Apollo rather than on SpaceX's own books. Leasing AI chips like aircraft is becoming a common way to fund the AI build-out, and it moves the risk onto private-credit investors.

Sources: SpaceX Form 8-K via StockTitan, Yahoo Finance (Reuters)

5. Australia's cyber agency warns online services may go offline unexpectedly as AI speeds up attacks

The Australian Signals Directorate (ASD) said in its annual cyber threat report that hostile governments and criminals are using AI to find weaknesses and attack faster, and that organisations should put "system security over system availability", even if that means taking services offline to fix them. ASD responded to 1,500 cyber incidents in 2025–26, up 20%, and warned outside organisations of more than 4,400 potential incidents, up 80%. Director-General Abigail Bradshaw said defenders need AI tools of their own to keep up; she has previously spoken about ASD having access to Anthropic's restricted Mythos model.

Source: The Guardian

6. Tata Communications will spend up to $450 million on a 10,000 km "AI super highway" of fibre across India

Tata Communications announced planned investment of $400–450 million to lay about 10,000 km of new long-haul fibre linking India's main data-centre and AI hubs, joined to its global network. The first phase covers the Mumbai–Pune–Hyderabad–Chennai corridor, with two more phases to follow. It is aimed at global cloud giants, AI "neocloud" firms, global capability centres and enterprises. CEO Ganesh Lakshminarayanan said data-centre-to-data-centre connectivity in India could become a market worth more than $1 billion, and that much of the expansion is backed by anchor customers.

Source: Tata Communications

7. Japan asks AI companies to disclose how they train their models and what kinds of data they use

Japan's Cabinet Office began applying a set of voluntary "principles" for generative AI companies on Monday, Yonhap reported, citing the Nikkei. Companies are asked to publish an outline of their training methods and the types of data used, answer rights holders who want to check facts before taking legal action over copyright, and answer users' questions about the copyright risk of AI output. The code is not legally binding; the government will publish a list of companies that sign up, with links to how each applies it. The minister for IP strategy said the guidelines would be reviewed if needed, which is read as a hint that they could become mandatory.

Source: Yonhap News

8. Study of 487 consultants: ChatGPT let them do data-science work beyond their skills, but they learned little

A randomised trial published in Nature Human Behaviour gave 487 management consultants at BCG access to and training in ChatGPT to solve data-science tasks outside their expertise, tasks ChatGPT could not solve alone. Those with ChatGPT scored higher than the control group on all three tasks, though not as well as real data scientists. But afterwards, without ChatGPT, they were no better at answering technical questions. The authors call AI an "exoskeleton": the boost lasts only while you keep using it. The study involved researchers from Boston University, BCG and OpenAI.

Source: Nature Human Behaviour

9. Microsoft starts rolling out Copilot Code, which builds work apps from plain-English requests

Microsoft has begun rolling out Code, one of the three parts of its revamped Copilot app announced last month, through its Frontier early-access programme. Users describe an app, tracker, dashboard or automation in ordinary language, and Copilot builds it, pulling missing details from their emails, meetings, documents and chats through Work IQ. Apps can run locally or be published to Microsoft's cloud and shared across a company, and bigger projects can be handed to developers as a GitHub pull request. Microsoft says it uses the same agent technology as GitHub Copilot.

Source: Digital Trends via Yahoo Tech

10. India's IT giants turn "policing" AI agents into a new business line

Mint reports that TCS, Cognizant, HCLTech and Tech Mahindra all now say managing and governing AI agents is a growing business, as companies that rushed to deploy agents worry they may go rogue or reach data they should not. Cognizant launched AI Nexus on 23 Oct, which gives each agent its own ID, permissions and memory, logs every task and keeps humans signing off on key decisions. TCS says an "agentic control plane" is becoming a top need for clients; HCLTech plans to buy Croatia's Robotiq.ai for agent orchestration; Tech Mahindra stresses red-teaming and traceability.

Source: Mint

11. New firm SWI Compute orders about $1 billion of Nvidia systems to build AI data centres in Europe, starting in Cambridge

SWI Compute, part of Amsterdam-listed SWI Group, came out of stealth on Monday. It has ordered about $1 billion of Nvidia systems and signed its first customer, a US AI inference platform. Its first large site, in the UK, starts at 30 megawatts with close to 7,000 of Nvidia's new Rubin GPUs and is expected online by end-2027, with the campus planned to grow to 530 MW and the company targeting 2.3 GW across Europe by 2029. The group says it spent six years securing 3.6 GW of grid power across Europe and the US, which it calls Europe's main AI bottleneck.

Source: PR Newswire

Sources are linked under each item.