The Daily Five · No. 9 · Wednesday 21 October

Today’s five stories ask who holds the controls as AI reaches the battlefield, the car, the cost of computing, the internet’s system of trust and the factory floor.

By the AI Story Front editorial team · About our editors

The UN moves to confront military AI. Can governments prevent machines from influencing life-and-death decisions?

A new international initiative aims to address the growing use of AI in defence, including nuclear-related decision-making. The challenge is whether countries competing for military advantage can agree on meaningful limits.

The United Nations Institute for Disarmament Research (UNIDIR) is preparing a new international forum addressing the military use of artificial intelligence, as governments face growing pressure to establish common safeguards.

The initiative is expected to begin on 22 October through UNIDIR’s Centre of Excellence on AI, Peace and Security.

According to reporting on the initiative, UNIDIR director Robin Geiss warned at a UN press conference that the expanding use of AI in military systems is reducing the time available for human judgement.

The concerns extend beyond autonomous weapons.

AI is increasingly being considered for intelligence analysis, surveillance, reconnaissance, early-warning systems and other activities that can influence military decisions.

Such applications may help defence organisations analyse information more quickly and recognise patterns across large datasets.

But greater processing speed does not necessarily produce better decisions.

An AI system may incorrectly classify a threat, fail to recognise important context or generate misleading conclusions that human operators struggle to assess under pressure.

The consequences could be particularly serious when automated assessments influence decisions involving nuclear weapons.

Geiss acknowledged the difficulty of determining precisely when the growing use of AI begins to undermine meaningful human control.

UNIDIR’s planned initiative is intended to bring together governments, defence technology companies, researchers and other organisations to examine these risks.

The institute has already published detailed guidelines for developing national strategies on AI in security and defence.

According to the guidelines, the Centre of Excellence is funded by the governments of France, Germany, Italy, the Netherlands, Pakistan, the Republic of Korea and Switzerland, and by Microsoft and a Geneva-based foundation.

The guidance addresses governance, technical testing, international law, institutional responsibilities and mechanisms for oversight and verification.

It also recommends cooperation with civil society and technology providers.

These recommendations recognise that military AI is not simply a technical question.

Defence institutions must decide which applications are acceptable, what evidence is required before deployment and who remains responsible when automated systems contribute to consequential decisions.

Yet international agreement is difficult because states also regard advanced AI as a potential source of military advantage.

Countries may recognise the dangers of uncontrolled automation while remaining reluctant to impose restrictions that competitors might not observe.

The UN initiative does not establish a binding international treaty or an independently enforceable regulatory framework.

Its immediate significance lies in creating a forum for examining risks, developing common approaches and encouraging greater cooperation.

Source: Associated Press

The Storyfront Perspective

Military AI raises a fundamental question about political power: who should have the authority to make decisions that can determine whether people live or die?

Technology may improve intelligence analysis, logistics and defensive capabilities. Faster identification of genuine threats could help prevent harm.

But military decisions take place within political and strategic contexts that cannot be reduced to the accuracy of a model.

An automated assessment may influence how commanders interpret an event long before a formal decision to use force is made.

This is why preserving human control requires more than keeping a person at the final stage of a decision.

That person must have sufficient information, time and institutional authority to question what the system recommends.

There is also a structural contradiction.

Governments are responsible for national security, but increasingly depend on private technology companies for models, computing infrastructure and specialist expertise.

Those companies may become influential actors in military systems without being subject to the same public accountability as elected governments.

The commercial interests of defence contractors and AI developers are not necessarily identical to the public interest in preventing conflict.

International cooperation therefore needs to address more than technical safety standards.

It must consider how AI changes military incentives, which institutions control its deployment and whether existing arrangements allow meaningful scrutiny.

Perhaps the most difficult problem is that governments have incentives to develop capabilities they simultaneously fear their rivals might acquire.

A credible international framework will need mechanisms that allow countries to verify compliance rather than rely exclusively on voluntary assurances.

The purpose of military AI governance should not simply be to make automated systems more reliable. It should be to prevent technological competition from weakening political responsibility for the use of force.

Gemini can now use connected Google services to answer questions during journeys. The feature promises convenience, but extending personal AI into vehicles raises questions about privacy, consent and control over everyday information.

Google has begun introducing Personal Intelligence to Gemini in cars, allowing its AI assistant to use information from connected services to provide more personalised responses.

According to Google’s announcement on 20 October, the feature can draw on Gmail, Google Calendar, Photos, Maps, YouTube and Search when users authorise the connections.

The system is intended to help drivers find information without searching through their phones.

Someone travelling to an airport could ask Gemini to find a flight confirmation and check the departure gate. Another user might request directions to an appointment without remembering the address.

Google says the assistant can also combine information across services, allowing more natural requests involving previous conversations, locations and personal preferences.

Gemini is already available through Android Auto and vehicles with Google built-in. Personal Intelligence is rolling out on Android Auto now and will reach cars with Google built-in in the coming months.

The initial rollout is available to eligible personal Google accounts for users aged 18 and over, but excludes the United Kingdom, European Economic Area, Switzerland and Nigeria.

Google says the feature is optional and that users can disconnect their apps through Gemini settings.

For cars with Google built-in, the company also says personalised responses are available only when the user’s phone is connected.

These controls address some immediate privacy concerns, although they do not establish how effectively the feature performs in all driving conditions.

The potential benefit is straightforward. Drivers may be able to retrieve information without handling their phones or navigating several applications.

However, an assistant that connects communications, photographs, travel history and personal preferences also brings together information previously held across different services.

The question is not only whether the system can protect that information, but how much access is necessary to complete individual tasks.

Source: Google

The Storyfront Perspective

Personalised AI may offer genuine benefits in a car, where reducing the need to use a screen could improve convenience and potentially reduce distraction.

But Google’s announcement also illustrates how personal information becomes more valuable when different services are connected.

A person might separately use Gmail, Maps and Photos without considering the combined picture those services create. An AI assistant capable of drawing information from all three can develop a much richer understanding of daily behaviour.

That creates a tension between convenience and autonomy.

Consent is important, and Google’s decision to make Personal Intelligence optional is a relevant safeguard. But agreeing to connect an account does not necessarily mean a person can anticipate every inference the system may draw.

Cars also create shared environments. Passengers may hear personalised responses or become part of conversations involving a service they have never agreed to use.

The distinction between access to information and control over information is therefore essential.

A useful assistant should allow people to restrict what it can retrieve, understand when personal information is being used and disconnect services without losing access to basic functions.

Google says those basic functions remain available without Personal Intelligence.

The broader challenge is ensuring that convenience does not make extensive data integration the expected condition of participation in everyday digital life.

IBM backs open-source AI infrastructure that could cut operating costs. Who will benefit?

Running advanced AI models is expensive, even after they have been developed. A new open-source release promises greater efficiency, but cheaper computing does not automatically mean cheaper AI for users.

A group of technology companies led by IBM, Red Hat, Google Cloud, CoreWeave and Nvidia has released llm-d 1.0, an open-source system designed to make large AI models cheaper and more efficient to operate.

The 20 October announcement marks a new stage for the project, which was established in 2025 and is now governed by the Cloud Native Computing Foundation.

Unlike developing or training a model, inference is the process of running it to produce responses. As companies introduce AI into everyday operations, those computing costs become increasingly important.

llm-d is designed to distribute requests across computing resources more efficiently, reducing unnecessary processing and improving how systems handle large numbers of simultaneous users.

The latest release introduces more advanced traffic management, support for multiple customers sharing infrastructure and features intended to keep services operating during maintenance.

It also improves support for AI agents, which may repeatedly access stored information and perform several actions before completing a task.

IBM says recent demonstrations using llm-d achieved five to ten times lower costs for certain complex agent workloads compared with commercial API pricing.

Other tests showed substantially higher processing capacity, including a demonstration supporting thousands of concurrent coding agents.

These results are tied to particular workloads, hardware configurations and pricing assumptions. They should not be interpreted as evidence that every organisation will achieve comparable savings.

The project has attracted more than 500 contributors from over 25 organisations, according to IBM.

Its open-source structure allows participating developers to examine and adapt the software, potentially reducing reliance on proprietary systems for managing AI workloads.

However, the software still requires access to computing hardware, technical expertise and suitable infrastructure.

For smaller organisations, those resources can remain expensive even when the software itself is freely available.

Source: IBM

The Storyfront Perspective

The cost of AI is becoming as important as its capabilities.

A company may be able to access a powerful model but still lack the financial resources to operate it at scale. This gives an advantage to organisations that control computing infrastructure and can distribute costs across millions of users.

Open-source technologies such as llm-d could help reduce that imbalance. More efficient use of hardware may lower barriers for smaller providers and make it easier to operate AI without relying exclusively on proprietary services.

But we should distinguish lower production costs from broader economic benefits.

If a large cloud company reduces its operating expenses, that saving may increase its profit margins rather than lower prices for customers.

There is also a contradiction worth examining. Several companies supporting this open-source project are themselves major suppliers of AI infrastructure. A more efficient software ecosystem could expand competition while simultaneously increasing demand for their hardware and cloud services.

This does not make the project less valuable. It shows why technological openness and economic competition are not necessarily the same thing.

The important measure will be whether organisations gain meaningful freedom to choose providers, control their systems and retain a greater share of the value AI creates.

Google blocks compromised security certificates. What happens when the internet’s system of trust fails?

A security researcher exposed weaknesses in certificates used to establish trusted connections with websites. Google responded rapidly, but the incident reveals how much internet security depends on organisations users rarely encounter.

Google has blocked a group of compromised digital certificates associated with eMudhra, an Indian certification authority, after a security researcher identified weaknesses that could allow attackers to impersonate trusted websites.

In a security report published on 20 October, the Chrome and Android security teams said they were alerted on 17 October to compromised private keys associated with eight subordinate certification authority certificates.

The problem involved insufficient randomness when the private keys were generated.

In cryptography, private keys must be sufficiently unpredictable to prevent outsiders from recreating them. Weaknesses in key generation can allow attackers to produce fraudulent certificates that appear legitimate.

Digital certificates underpin HTTPS, the technology used to establish secure connections between browsers and websites.

When a browser recognises a certificate as trustworthy, users see familiar indicators that the connection is protected.

But that trust relies on a network of certification authorities responsible for issuing and managing credentials.

If an authority’s keys are compromised, attackers may be able to create certificates that falsely appear to authenticate websites.

Google says Chrome engineers verified the researcher’s findings within two hours and deployed protections blocking the affected keys.

Android devices subsequently received equivalent protections through Google Play Services.

According to Google, the affected certificates had not been recorded in the Certificate Transparency logs required by Chrome before the blocking action. This limited the possibility of fraudulent certificates being accepted by clients enforcing those transparency requirements.

The company also announced plans to remove the root certificate associated with the affected infrastructure from Chrome’s trusted certificate store.

That change, expected to reach the relevant Chrome versions from 27 October, could affect legitimate websites using certificates issued under the same authority.

Those website operators will need to obtain replacement certificates through another trusted provider.

Google says it is coordinating with other browser developers and infrastructure providers while reviewing the continued inclusion of eMudhra’s remaining certification authorities.

The response demonstrates both the importance and the complexity of digital trust.

Removing compromised credentials can protect users, but doing so can also disrupt legitimate services dependent on the affected provider.

Source: Google

The Storyfront Perspective

The internet’s security depends on an important distinction between technical trust and institutional trust.

Encryption can protect communication, but users must also trust the organisations that certify the identity of the websites they visit.

Most people cannot inspect this system or choose which certification authorities their browsers recognise. Those decisions are largely made by technology companies, standards organisations and specialised infrastructure providers.

That concentration of authority has practical advantages.

Google’s ability to distribute protections quickly across Chrome and Android helped contain a potentially serious security weakness.

Yet the same incident demonstrates how failures within a relatively obscure company can create risks extending far beyond its own customers.

It also raises questions about oversight.

What standards govern the generation and protection of cryptographic keys? How frequently are certification authorities independently assessed? And what remedies are available when a provider’s failure disrupts websites that relied on it?

The issue is not whether large technology companies should intervene. In this case, intervention was necessary.

The deeper question is how a global communications system can remain secure without depending excessively on the competence and decisions of a relatively small number of private institutions.

Trust should be supported by verifiable standards and effective accountability, rather than simply assumed because an organisation has been granted authority.

America leads in AI but falls behind in industrial robotics. Who will benefit from automation?

A new US commission report argues that leadership in artificial intelligence has not translated into leadership in manufacturing. The findings raise questions about investment, industrial policy and whether workers will share in the gains from automation.

The United States risks falling behind in industrial robotics despite its strength in artificial intelligence, according to a new report from the National Security Commission on Robotics for Advanced Manufacturing.

The commission’s October 2026 preliminary report identifies a growing gap between America’s ability to develop advanced AI and its capacity to deploy robotics across factories and industrial production.

China, the report argues, benefits from a much larger installed base of industrial robots and extensive manufacturing supply chains.

The US faces a different problem: manufacturers are reluctant to invest in robots until they become cheaper and more reliable, while robotics companies need greater demand to reduce costs and improve performance.

The commission describes this as a cycle that cannot easily be resolved by market competition alone.

Smaller manufacturers face particular difficulties. Purchasing robots requires significant capital, but successful deployment also depends on specialised engineers, integration expertise and changes to existing production processes.

The report identifies shortages of skilled workers capable of installing, maintaining and operating advanced robotic systems.

It also warns that American robotics companies face financing problems because investors often expect returns more quickly than complex hardware businesses can deliver.

These challenges matter as AI makes robots more adaptable.

AI-assisted programming and improvements in machine perception could allow machines to perform tasks that previously required expensive customisation.

The potential benefits include improved workplace safety, greater productivity and more competitive domestic manufacturing.

But those outcomes are not automatic.

The commission acknowledges that automation can displace some jobs, transform others and create new forms of employment. Its effects will differ between industries, regions and workers.

The report calls for better coordination of industrial investment, workforce training and robotics deployment. Its final recommendations are expected in March 2027.

Source: Special Competitive Studies Project

The Storyfront Perspective

The report exposes a contradiction in the American approach to technological leadership.

The United States has produced some of the world’s most powerful AI companies. Yet technological innovation does not automatically translate into stronger manufacturing, better employment or higher living standards.

The report itself recognises that market incentives have failed to produce sufficient investment in robotics.

This raises a question about the relationship between public policy and private economic power.

Governments may need to support industrial infrastructure, skills and investment to make automation viable. But public intervention should not simply subsidise private productivity gains.

If companies use publicly supported technologies to reduce labour costs, who benefits from the increased efficiency?

Workers could gain safer conditions, better tools and opportunities to develop new skills. They could also experience displacement, reduced bargaining power or greater pressure to produce more with fewer colleagues.

These outcomes depend on decisions about ownership, employment and how productivity gains are distributed.

There is also an international dimension. Competition with China is presented primarily as a question of national industrial leadership, but a country’s technological success does not necessarily mean that prosperity is shared equally within it.

The measure of successful automation should therefore extend beyond the number of robots installed or the competitive position of American manufacturers.

It should include the quality of employment, the distribution of productivity gains and the ability of workers to influence technological changes affecting their livelihoods.

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

The UN moves to confront military AI. Can governments prevent machines from influencing life-and-death decisions?

Gemini brings personalised AI into cars. Who controls the information passengers share?

IBM backs open-source AI infrastructure that could cut operating costs. Who will benefit?

Google blocks compromised security certificates. What happens when the internet’s system of trust fails?

America leads in AI but falls behind in industrial robotics. Who will benefit from automation?

Five AI stories that matter, and what they mean. Every weekday.