AI Daily Brief: 26 June 2026

26 June 2026

Today's AI news is about control. Governments are slowing frontier releases, employers are trying to rework labour markets, and AI vendors are pushing agents and smaller edge models into mainstream business workflows.

OpenAI slows GPT 5.6 rollout after White House safety pressure

OpenAI will reportedly release GPT 5.6 first to a limited group of close partners instead of pushing it straight to the public. Sam Altman is said to have told staff that access will be approved customer by customer during the preview period.

The request reportedly came from the Office of the National Cyber Director and the Office of Science and Technology Policy, after a June executive order asking certain AI companies to submit new models for government testing before public release.

For UK businesses, the signal is clear: frontier model access is becoming a policy question as much as a product question. Procurement teams should plan for staged availability, customer restrictions and extra risk reviews around high capability cyber models.

Our take: This is the new reality of frontier AI. The most powerful models are no longer simple software launches. They are becoming controlled infrastructure, and businesses that depend on them need contingency plans for delayed access, jurisdiction limits and sudden policy intervention.

AI giants back $500 million worker transition push

Amazon, Anthropic, Microsoft and the OpenAI Foundation are among the anchor partners backing RAISE US, a bipartisan non-profit led by former US Commerce Secretary Gina Raimondo and former Indiana Governor Eric Holcomb.

The initiative aims to pilot employer incentives, wage insurance, career navigation and new training models for workers affected by AI-driven labour disruption. Reporting puts the initial backing at $500 million, with wider ambitions to shape national workforce strategy.

For business leaders, the important point is not the US geography. It is that AI vendors are now acknowledging workforce disruption as a structural issue, not a communications problem. UK employers should expect sharper questions from staff, unions and regulators about retraining before automation programmes scale.

Our take: The initiative is useful, but it also exposes the gap. Companies cannot treat AI as a headcount reduction machine and then bolt on training after the fact. Workforce planning needs to be part of the AI business case from day one.

OpenAI says Codex usage is moving beyond developers

OpenAI says agentic AI usage is growing quickly inside and outside the company. Its researchers report that active Codex users grew more than fivefold in the first half of 2026, with the fastest growth coming from people outside the original software developer audience.

Inside OpenAI, 97.9 percent of employees now use Codex, up from about 40 percent in August 2025. External organisational usage is reported at 17.3 percent, while non-developer Codex use has risen 137 times for individuals and 189 times for organisational users.

The practical message for UK firms is that coding agents are no longer just engineering tools. Legal, recruiting, operations and finance teams are beginning to use them for automation, data transformation and analysis, which changes where technical governance needs to sit.

Our take: The useful question is not whether non-developers can use agents. They already are. The governance question is whether the business has review processes, access controls and technical escalation paths for work that looks non-technical until it changes data, code or workflows.

Amazon commits another $13 billion to AI and cloud infrastructure in India

Amazon has announced another $13 billion for AI and cloud infrastructure in India, focused on extra AWS datacentre capacity in Mumbai and Hyderabad. The commitment sits inside a broader five-year India investment pledge of more than $48 billion.

The company says the infrastructure will support startups, enterprises and government organisations with custom AI chips, managed AI services, cloud technologies and developer tools. Amazon has also said its total commitments in India between 2010 and 2030 exceed $88 billion.

This matters for UK firms because AI capacity is increasingly geopolitical. Where compute is built affects latency, data residency, supply chain resilience and vendor leverage. Multinationals should map where their AI workloads actually run, not just which vendor brand appears on the invoice.

Our take: AI strategy is becoming infrastructure strategy. Businesses that treat cloud regions as a technical detail will miss the commercial and regulatory consequences of where compute capacity is being concentrated.

Elastic cuts about 7 percent of workforce while leaning into AI automation

Elastic is cutting about 7 percent of its workforce as part of a restructuring tied to AI automation, simpler team structures and faster decision-making. Reports put the affected headcount at roughly 281 to 300 workers.

The company expects cash restructuring costs of $22 million to $25 million, mainly for severance, while still saying it plans to keep hiring in strategic areas and grow overall headcount this fiscal year. Elastic reported Q4 FY 2026 revenue of $451 million, up 16 percent year on year.

The pattern is becoming familiar: AI is used to reduce complexity in some functions while companies continue hiring elsewhere. For UK businesses, this means AI adoption should be measured by capability shifts, not just net headcount changes.

Our take: The easy narrative is that AI replaces jobs. The more useful management lesson is that AI changes the shape of the organisation. Some roles shrink, others grow, and the dangerous bit is failing to tell people which is which early enough.

Liquid AI releases 230 million parameter model for edge deployment

Liquid AI has released LFM2.5-230M, an open-weight model designed for fine-tuning, tool use, data extraction and edge deployment. The company says the model can run from cloud GPUs down to low-cost CPUs.

Liquid reports 213 tokens per second decode speed on a Galaxy S25 Ultra and 42 tokens per second on a Raspberry Pi 5. VentureBeat reports the model scored 43.26 on the BFCLv3 tool-use benchmark, ahead of IBM Granite 4.0-350M at 39.58.

For businesses, this points to a quieter but important trend: not every useful AI workload needs a frontier model or a cloud round trip. Private, local, cheap inference will matter for field work, devices, regulated environments and low-latency operations.

Our take: Small models are moving from novelty to deployment option. The sensible architecture in 2026 is not one giant model for everything. It is routing work to the cheapest capable model, including on-device models where privacy, cost or latency matter.

Chinese security firm claims AI bug finder rivals Claude Mythos

Qihoo 360 says it has built an AI vulnerability finder that can rival Anthropic's Claude Mythos. Reporting says the company is positioning the system as a domestic answer to advanced AI cyber models and as a deterrent against weaponised frontier systems.

Quartz reports that 360's Tulongfeng system found more than 3,400 software vulnerabilities. The Register reports that CEO Zhou Hongyi presented the tool as part of a wider argument that automated vulnerability discovery is becoming a strategic cyber capability.

For UK organisations, the commercial implication is direct. AI-assisted vulnerability discovery will help defenders, but it will also raise the tempo for attackers. Patch speed, asset visibility and software supply chain controls need to improve before automated discovery becomes routine.

Our take: This is the cyber arms race version of AI adoption. The winners will not be the organisations with the most impressive demos. They will be the ones that can verify, patch and govern faster than automated tools can find the next weakness.

Quick Hits

• OpenAI updated GPT 5.5 Instant for better intent understanding, complex constraints, shopping and local recommendations.
• The Register says Amazon's India AI and cloud investment forms part of a wider $48 billion five-year pledge.
• Business Journals reports Elastic's 7 percent workforce cut affects roughly 281 workers.
• Liquid AI's LFM2.5-230M weights are available on Hugging Face for developers to test and fine-tune.

Frequently Asked Questions

How often is the AI Daily Brief published?

Every morning at 7:30am UK time, covering the previous 24 hours of AI news from over 30 sources.

How are stories selected?

UK-relevant stories are prioritised first, then by business impact and practical implications for UK organisations adopting AI.

Why should business leaders follow AI news?

AI is moving faster than any technology in history. Staying informed is essential for making smart decisions about AI investment, adoption, and governance.