Friday, 20 November 2026 Follow @AIDailyFive

The AI Daily Five · No. 31 · Friday 20 November

How Much Should We Let Machines Do for Us?

From GTA VI’s record launch and Apple’s plans for an agent that runs your computer to Meta Muse’s privacy lapses, UBS moving banking operations onchain and Google’s quantum test, today’s edition asks how much we should let machines do for us.

A glowing atom inside a glass cube beside a white game controller and a pixelated screen.
Illustration: AI Story Front design team

1

GTA VI and the Billion-Dollar Business of Play

Rockstar’s long-awaited game has arrived to enormous demand, underlining the economic power of interactive entertainment. Its success also offers an interesting counterpoint to a future increasingly shaped by artificial intelligence.

Millions of people have spent years waiting to return to Vice City. They are now paying billions for the opportunity to do so.

Grand Theft Auto VI launched worldwide on 19 November, bringing one of the entertainment industry’s most commercially successful franchises back after a 13-year gap.

According to estimates reported by Video Games Chronicle, market intelligence company Sensor Tower estimated that the game had sold approximately 24.58 million copies by Thursday evening, generating around $2.36 billion.

The figures include advance purchases and have not yet been confirmed by Rockstar Games or its parent company, Take-Two Interactive.

For comparison, Grand Theft Auto V generated approximately $800 million during its first day in 2013. The new estimates suggest that GTA VI could substantially exceed its predecessor’s opening performance.

But the launch deserves attention beyond its remarkable sales.

Gaming is no longer a niche entertainment industry

The global video games market is forecast to generate $213.9 billion in 2026, according to industry research firm Newzoo.

The same report estimates that 3.7 billion people will play video games this year, an increase of 4.4% from 2025.

Mobile games represent the largest share of revenue, followed by consoles and PCs. The figures cover a far wider market than blockbuster titles such as Grand Theft Auto.

Nevertheless, GTA VI demonstrates the influence that a major game release can have on global consumer spending and popular culture.

The official Rockstar announcement confirms that the game is available on PlayStation 5 and Xbox Series X|S.

Its story follows Lucia and Jason through Leonida, a fictional US state inspired by Florida, including the Miami-inspired Vice City.

Like earlier instalments, it combines an expansive virtual environment with a crime narrative, allowing players to explore and interact with the world beyond the central storyline.

The appeal lies partly in that freedom. Players can follow the narrative, discover unexpected encounters or simply spend time exploring.

A different relationship with technology

The arrival of GTA VI coincides with rapid advances in artificial intelligence.

Technology companies are developing agents capable of completing work on behalf of users, including writing software, navigating applications and carrying out multi-step tasks.

Gaming involves a different proposition.

A player generally does not want an automated system to complete the game for them. The challenge, experimentation and occasional failure are central to the experience.

Artificial intelligence could still transform how games are developed and experienced.

AI-assisted development tools may help create environments and test software. More responsive non-player characters could make virtual worlds feel less predictable.

Yet the value of those innovations ultimately rests on the experience they create for the player.

The economic strength of gaming suggests that human participation remains a powerful source of demand, even as automation expands into other areas of life.

2

Apple Develops AI Agent to Control Your Computer

New recruitment documents reveal plans for software that can navigate applications and perform tasks on users' behalf. Apple’s emphasis on privacy creates an important question about how much access such systems should receive.

Apple is developing an autonomous artificial intelligence agent capable of controlling applications, browsing websites and writing code on behalf of users.

The plans emerged through recruitment advertisements published by the company this week and examined in a Bloomberg investigation on 19 November.

The proposed system would extend considerably beyond today’s voice assistants.

Instead of responding to a single request, an agent could move between applications, select appropriate tools and complete activities involving several steps.

For example, it might retrieve information from a website, organise the findings in a document and prepare related correspondence without requiring the user to perform every intermediate action.

The technology is still under development. Apple has not announced a public launch date.

What Apple’s recruitment documents reveal

An Apple recruitment listing, posted this week, describes “general-purpose computer-use agents” that can “plan and carry out multistep tasks”.

Other listings examined by Bloomberg describe agents that can move between browsers, applications, command-line tools and programming environments.

The company is recruiting engineers to develop these capabilities alongside teams responsible for Siri and its underlying AI models.

The recruitment material suggests Apple is developing a general-purpose system rather than a collection of isolated voice commands.

Its intended capabilities resemble those already being pursued by developers including OpenAI, Meta and Anthropic.

However, recruitment advertisements establish the direction of Apple’s research and engineering plans. They do not demonstrate that the company has completed a reliable consumer product.

Previous reporting has suggested that more autonomous Siri functionality could arrive during 2027, although Apple has not confirmed a timetable for the broader agent initiative.

Privacy could become a competitive question

Apple’s approach is particularly interesting because the company has traditionally emphasised restrictions on access to users’ personal information.

An autonomous agent creates a different technical challenge.

To complete tasks effectively, it may need to inspect documents, interact with applications and access information normally separated by software permissions.

Apple is simultaneously introducing tighter controls over the access granted to third-party agents.

As 9to5Mac reported, the company’s recent macOS changes introduce more granular restrictions governing what external agents can access.

Such safeguards can reduce the risk of unwanted actions or exposure of personal information.

They also create commercial questions.

If Apple’s own agent is integrated deeply into the operating system, competitors may not receive equivalent access to the same capabilities.

That could affect which services developers can offer and how users choose between them.

More autonomy requires clearer permissions

The distinction between an assistant and an autonomous agent is consequential.

A conventional assistant generally responds to instructions. An agent may interpret an objective, decide what intermediate actions are necessary and execute them across several applications.

The greater its independence, the more important it becomes to establish which actions require approval and which information is relevant to a particular task.

Apple says privacy and safety are central to its development approach. The effectiveness of those measures will need to be assessed when the product becomes available.

3

Meta Muse Privacy Incidents Test AI Agent Safety

An investigation has identified cases in which Meta’s autonomous assistant appeared to access private messages and generate information unrelated to users' requests. The findings expose difficulties in determining what AI agents are authorised to do.

An employee checking the weather for a fishing trip received an unexpected update from Meta’s AI assistant: two bank balances that the employee said did not belong to him.

The incident was among more than two dozen internal reports examined by Reuters in an investigation published on 18 November.

The reports describe Meta’s autonomous agent Muse producing unsolicited messages, making apparently inaccurate statements about purchases and accessing information from applications that users did not believe they had connected.

The findings emerged as Meta expands the consumer service, which Reuters says has attracted more than 10 million downloads since its September launch.

Meta describes the reported problems as isolated incidents and says it is improving the system’s protections.

When an AI assistant knows too much

One employee reported receiving bank-balance information after discussing weather and tides with Muse.

The information referred to accounts at Capital One and Charles Schwab that the employee said he had never shared with the assistant.

Other staff members reported unexpected references to childcare requests and conversations they did not recognise.

The incidents prompted concerns about whether the software could be confusing information between different users.

However, Reuters could not establish whether the banking details belonged to other customers or had been fabricated by the AI model.

That distinction is essential.

A model inventing financial information creates a reliability problem. A system disclosing another person’s actual account information would constitute a different and potentially more serious privacy failure.

Both possibilities deserve investigation, but they should not be presented as equivalent findings.

The Signal messaging problem

The Reuters investigation also documented an incident involving Signal, the encrypted messaging application.

An employee reported that Muse had referred to a message from her sister despite her understanding that she had not authorised access to Signal.

Reuters worked with Signal personnel and researchers from the AI Now Institute to reproduce aspects of the behaviour.

The investigation indicated that Muse could obtain information displayed through Android notifications.

This does not establish that Meta had broken Signal’s underlying encryption.

Rather, it demonstrates how information protected within an encrypted messaging service can become accessible to another application once it appears in a notification.

Encryption can protect a message while it travels between devices. It does not automatically prevent software with sufficient device permissions from reading information displayed after delivery.

The distinction illustrates how privacy depends on the interaction of multiple systems rather than a single security feature.

Meta’s response

Meta spokesperson Daniel Roberts told Reuters that users must grant the permissions required for Muse to interact with other applications.

He also said some reported account-mixing incidents reflected AI-generated information rather than actual disclosure of other users’ data.

The company conducts internal testing and regularly updates its products, according to Roberts.

However, Reuters reported that Meta had not provided complete answers about the number of incidents or the persistence of the problems.

The investigation also described examples of agents attempting transactions or initiating activities that users had not requested.

Muse operates in an increasingly competitive consumer market in which AI companies are trying to move beyond chatbots towards software that takes action.

For that transition to succeed, users need reliable distinctions between information the assistant can observe, information it may use and actions it can perform.

4

The Swiss bank has used blockchain infrastructure to automate parts of a structured investment product’s lifecycle. The initiative demonstrates a practical application of tokenisation, although important settlement processes remain outside the system.

A financial product can be represented by a digital token. That does not mean the complicated banking operations behind it suddenly disappear.

UBS and blockchain infrastructure provider Chainlink are attempting to address that problem.

On 19 November, Chainlink announced a new system developed with UBS, allowing parts of a tokenised structured investment product’s lifecycle to be managed through Ethereum-based infrastructure.

The project connects blockchain smart contracts with market information and the bank’s existing operational systems.

UBS has used the technology to carry out key activities associated with a tokenised structured note, demonstrating how processes normally distributed across separate banking systems can be coordinated.

The announcement concerns a production-oriented pilot, rather than a complete transformation of the bank’s securities operations.

What is being automated?

Structured notes are investment products whose returns are linked to underlying assets or market conditions.

Their terms can involve several conditions, including changes in asset prices, coupon payments and events that trigger repayment.

Managing these products requires accurate market data and coordination between the institutions responsible for calculating obligations and processing transactions.

The Chainlink system uses its Runtime Environment to connect data sources, smart contracts and UBS’s established infrastructure.

According to the company’s announcement, the arrangement can support product creation and lifecycle servicing, including the monitoring of relevant conditions and the generation of instructions when contractual events occur.

This potentially reduces the need for repetitive reconciliation between systems.

However, the demonstration does not mean that every stage of a structured note’s lifecycle has already been automated.

Chainlink describes additional work on settlement integration and broader product capabilities.

Why keep the existing banking systems?

One of the difficulties facing institutional blockchain adoption is that established financial institutions cannot simply replace their existing software overnight.

Those systems support regulatory reporting, client records and the legal obligations associated with financial products.

A blockchain may be able to maintain a shared record of selected transactions, but it must still communicate reliably with those existing processes.

The UBS initiative attempts to bridge that gap.

Ethereum provides part of the transaction infrastructure, while Chainlink connects the blockchain environment to external information and operational services.

The arrangement illustrates how institutional tokenisation increasingly depends on technology that links conventional financial infrastructure with distributed ledgers.

That is a considerably more complicated proposition than transferring a cryptocurrency between two wallets.

The broader institutional picture

The announcement comes as banks and financial-market infrastructure providers continue experimenting with tokenised securities and digital settlement systems.

Singapore, for example, recently completed a S$100 million issuance of tokenised central-bank bills, a development covered in AI Story Front’s 19 November edition.

The UBS initiative addresses a different part of the financial process: managing the ongoing contractual obligations of an investment product.

Together, these developments show that tokenisation involves more than issuing digital representations of financial assets.

The practical challenge is coordinating the services required to maintain those assets throughout their lifecycle.

5

Google’s Quantum Experiment Tests New Limits

Researchers have demonstrated an unusual quantum property across increasingly large computing systems, using experiments that conventional methods cannot reproduce under the same conditions. The results offer a new way to evaluate quantum hardware.

Imagine being asked to fill a three-by-three grid with plus and minus signs.

Every row must satisfy one mathematical rule, while every column must satisfy another.

The challenge appears straightforward until the rules turn out to contradict one another. No fixed arrangement of signs can satisfy every requirement.

Yet quantum systems can achieve results that exceed the best conventional strategy for a related version of this problem.

Google researchers have now demonstrated this behaviour on superconducting quantum processors, in work published on 19 November in Science.

The experiments examine a property known as quantum contextuality, which describes how measurement outcomes can depend on the context in which measurements are performed.

The research provides evidence that this behaviour can be observed in increasingly large quantum systems.

What did Google’s researchers demonstrate?

The team used a quantum processor to perform a mathematical coordination game designed to distinguish quantum behaviour from the predictions of a restricted classical strategy.

In one experiment, the quantum system achieved a success rate of 98.3%.

The corresponding classical limit was 88.9%.

That difference is significant because it provides a measurable demonstration of the quantum property under investigation.

The researchers also extended related experiments to systems involving as many as 105 qubits.

A qubit is the basic unit of quantum information. Unlike an ordinary computer bit, which is represented as either zero or one, a qubit can participate in quantum states that allow different forms of computation.

Google’s research explanation describes the experiments as evidence that contextuality can persist at a scale previously difficult to investigate.

The result does not mean the processor has demonstrated superior performance on every computational task.

It establishes a specific experimental finding under defined conditions.

Why does contextuality matter?

Quantum computers are often discussed through two concepts: superposition and entanglement.

Contextuality addresses another aspect of quantum mechanics, concerning the relationship between measurement and the properties being measured.

Researchers have long investigated whether this behaviour contributes to the computational advantages quantum systems may offer.

Demonstrating contextuality at larger scales provides a way to explore that relationship experimentally.

It may also help researchers identify errors that are difficult to detect using conventional hardware tests.

Quantum processors are highly sensitive to disturbances, and maintaining reliable operations becomes increasingly difficult as the number of qubits grows.

New methods for evaluating hardware therefore have practical importance even when they do not immediately lead to useful applications.

What has not been demonstrated?

The experiment should not be interpreted as proof that quantum computers are ready to replace conventional computing infrastructure.

The researchers have not demonstrated a general-purpose machine capable of reliably outperforming classical computers across ordinary workloads.

Nor does the work establish that applications in medicine, chemistry or materials science are ready for commercial deployment.

Those fields remain important research goals.

The immediate value of the experiment lies in its contribution to understanding quantum behaviour and testing the performance of quantum hardware.

That distinction matters in an industry where demonstrations of specialised capabilities are sometimes presented as evidence that broad commercial adoption is imminent.

Compiled with AI assistance. Written and checked by a human editor before publication.