The AI Daily Five · No. 20 · Thursday 5 November
Who Can We Trust With AI’s Answers?
From Britain’s £6 million for local news and Claude’s protein research to an SEC position on Canton Coin, Google’s health AI tests and Siri’s opening to developers, today’s edition asks who checks the claims.

1
Can £6 Million Save Britain’s Local Journalism?
Fifty organisations will receive public funding to strengthen local news, including projects using AI to support reporting. The grants address digital transformation, but also expose the financial pressures facing independent publishers.
The UK Government has awarded £6 million to 50 organisations through its Local News Fund, supporting community journalism, digital publishing and technology projects across England and Wales.
Announced on 5 November during Journalism Matters Week, the funding forms part of the government’s Local Media Action Plan. Most grants will go to local news providers, while seven organisations will develop infrastructure intended to benefit the wider industry.
According to the government announcement, publishers will receive grants of up to £125,000, while selected infrastructure projects can receive up to £275,000.
The recipients include community newspapers, independent digital publications and local radio stations. Projects range from developing sustainable membership models to expanding reporting capacity and improving access to council information.
Several grants specifically involve artificial intelligence.
The Hackney Citizen’s publisher will receive £89,667 for a platform combining AI-assisted investigations, data journalism and reader revenue. Newsquest has secured £125,000 for a London project incorporating AI transcription, article drafting and multimedia production, subject to journalistic review.
The National Council for the Training of Journalists will receive £191,880 to develop industry guidance and training on responsible AI use in local newsrooms.
Other projects focus on improving original reporting without necessarily relying on AI. They include funding to recruit journalists, develop community participation and strengthen financial relationships with readers.
The government says the Local News Fund is intended to support long-term sustainability, particularly as audiences move online and get their news from more sources.
A further £1 million has been earmarked for local news deserts in 2027, with up to £5 million more potentially available for wider industry support.
The National Union of Journalists welcomed the initiative, but said many members remain deeply worried about their livelihoods. It added that it will monitor the use and impact of the grants for “responsible” AI projects.
The funding has been awarded, but its impact on newsroom finances and the quantity of original reporting has yet to be demonstrated.
2
Can Claude Help Solve a 40-Year-Old Mystery?
Anthropic researchers have reconstructed a previously unknown protein structure built from a protein of a bacteria-infecting virus. The work offers evidence of AI contributing to laboratory research, although the biological purpose of these particles remains unexplained.
Scientists at Anthropic have reconstructed the three-dimensional structure of a vault-like protein particle associated with a bacteriophage, a virus that infects bacteria. Their findings, announced on 4 November, offer a new way to investigate molecular structures whose purpose has remained uncertain for decades.
Vault particles were first identified in 1986. These large, hollow protein structures occur in many complex organisms, including humans, but scientists have yet to establish their precise biological function. Earlier research had identified related proteins in bacteria and viruses without confirming that they assembled into complete particles.
Anthropic’s researchers used Claude to search approximately 15 billion predicted protein sequences, design DNA constructs and assist with laboratory procedures. The team then used cryo-electron microscopy to reconstruct a particle made from 96 copies of a single protein.
The company described the research as an example of AI participating throughout the scientific process. Researchers directed the experiments, while Claude helped analyse information, propose methods and process microscopy data.
One practical result was a substantial reduction in laboratory processing time. Anthropic reports that a purification process normally taking approximately three days was completed in around 90 minutes, while an initial structural model was ready after 34 hours.
The team also modified the protein to investigate how the particles assemble. Some alterations prevented the structure from closing, while others produced larger particles, offering clues about how these molecular complexes might have evolved.
The biological mystery, however, remains unresolved. Scientists still do not know what vault particles do, although possible applications involving molecular delivery have been discussed. So far, the team has produced the particles only in E. coli, and has not tested whether they form in the bacteria and viruses where the proteins occur naturally.
Anthropic has released a technical report and submitted a preprint to bioRxiv. The findings have not yet completed conventional peer review, and the reported efficiency improvements come from the company’s own laboratory. Anthropic also reports Claude’s mistakes, including a protocol error that set the project back two weeks and a flawed early map. The vault’s co-discoverer, Leonard Rome, advised on the work as a paid consultant.
3
Is Canton Coin a Commodity or a Security?
SEC staff have indicated that specific token-minting activities on the Canton Network would not trigger an enforcement recommendation. The decision offers regulatory clarity, but only under the circumstances presented.
The US Securities and Exchange Commission’s Division of Corporation Finance has issued a no-action response concerning Canton Coin, the digital token used by the institutional blockchain network Canton.
In its 3 November regulatory response, SEC staff indicated that they regard Canton Coin as a digital commodity under the interpretation cited in the decision.
The response concerns the token’s role in operating the Canton Network, an infrastructure designed to support financial applications using distributed ledger technology.
Canton Coin is used within the network’s incentive and operating arrangements. Unlike tokens marketed primarily around expectations of returns from a company’s management, its stated function is connected to network activity.
The SEC staff relied on representations that the coin derives its value from the operation of the network and market supply and demand, rather than expectations of profits generated by a central management team.
The response also cited the representation that the Canton Network has no central party as defined by the relevant regulatory interpretation.
On that basis, staff said they would not recommend enforcement action if Canton Coin were minted programmatically by authorised network participants without registration under specified provisions of US securities law.
This does not amount to a general exemption for all cryptocurrencies. The staff explicitly stated that the response depends on the facts presented and expresses a position about enforcement recommendations rather than a binding legal conclusion.
Different arrangements for token issuance, management or investor expectations could produce a different regulatory assessment.
The development matters for institutional blockchain projects because uncertainty over whether a token qualifies as a security can complicate network operations and participation.
Canton’s regulatory treatment may also interest other infrastructure providers seeking to distinguish tokens used for network functions from conventional investment products.
4
How Safe Are Google’s Medical AI Answers?
Google has published its approach to evaluating AI responses to health questions, combining automated benchmarks with medical expertise. The framework addresses accuracy and safety, although real clinical reliability requires evidence beyond company testing.
Google has published a framework for assessing the quality of artificial intelligence responses to health questions, addressing an area where inaccurate or misleading information can have serious consequences.
Released on 4 November, the approach evaluates medical AI systems across three dimensions: factual accuracy, safety and usefulness.
The company says the framework was developed with health experts, including Elizabeth McGlynn and Dana Gelb Safran, and is described in a new white paper.
According to Google’s published methodology, its evaluation process combines structured benchmarks, automated assessment tools and reviews by human experts.
The aim is to examine how models respond to common health questions as well as situations involving greater risks to patients.
Measuring accuracy alone is insufficient in healthcare. An answer may contain correct information but still be unsafe if it provides inappropriate advice for a particular situation or fails to recognise when professional medical attention is necessary.
Likewise, an answer that avoids obvious errors may be of limited value if it is difficult to understand or does not address the user’s actual question.
Google’s framework combines three methods across these areas, rather than relying on a single benchmark score. The accompanying white paper says this should help bridge the gap between benchmark results and real-world performance.
The company says it uses the approach internally when developing health-related AI capabilities and hopes publishing the methodology will help other researchers and developers.
However, the announcement does not establish that all Google AI systems are clinically safe in every deployment. The blog post describes the method, but does not report how Google’s models scored. Evaluations conducted under selected conditions cannot account for every medical situation, language difference or individual patient’s circumstances.
Independent testing, transparency about limitations and appropriate clinical oversight remain important, particularly where AI-generated information could influence treatment decisions.
The framework represents a development in how medical AI is assessed, rather than a newly approved clinical tool.
5
Can Siri Become the Gateway to Every App?
Apple is allowing developers to integrate their applications with its conversational assistant through a new preview. The change could simplify everyday tasks while giving the company greater influence over how customers access digital services.
Apple has opened a developer preview allowing third-party applications to integrate their functions directly with Siri AI, extending the assistant’s ability to perform actions across software services.
Announced on 3 November, the programme is available through beta versions of Apple’s operating systems, including iOS 27.3, iPadOS 27.3 and macOS 27.3.
The company says developers can use the new capabilities to make their applications accessible through natural-language requests. A user could eventually ask Siri to complete an action without navigating the application’s conventional interface.
Apple’s developer announcement describes examples involving deliveries, bookings and other activities carried out through connected services.
The integration builds on Apple’s existing App Intents framework, which allows applications to expose specific functions to the operating system.
The new preview extends that approach to a more conversational assistant capable of understanding requests and connecting them with relevant application actions.
Apple says the programme is intended to allow developers to experiment before broader customer availability in the coming months. App Store submissions supporting the new capabilities are expected to open in spring 2027. Apple says more details on terms and usage policies will follow.
The announcement does not mean every application already supports the new functionality. Developers must implement the relevant interfaces, while availability also depends on operating-system versions and compatible devices.
Apple says Siri AI currently supports several languages, including English, French, Japanese and Spanish.
However, some functionality remains restricted by geography. Apple says Siri AI is currently unavailable in the European Union on iOS, iPadOS and watchOS, and says it is working with the EU to find a path forward.
The company has also indicated that some server-based features, including Siri AI, are subject to daily usage limits, with expanded access to be offered for a fee in the future.
The development could change how users discover services and interact with applications, particularly if conversational assistants replace some conventional navigation.
It also places Apple in a position to shape the relationship between developers and customers through the operating system.
The AI Daily Five