Latest UK & Global AI News

'Agent Smith' has entered Googleplex, and it is reportedly so popular that the company has restricted it The Times of India Google gives some employees Agent Smith to free up time NewsBytes Google employees have a new AI tool called 'Agent Smith.' It's so popular that access got restricted. Business Insider Google employees now using AI agent to automate tasks, working using phone India Today Google’s Agent Smith AI is blowing up internally: Here is what it can do digit.in

View Full Article

The danger is not that machines are becoming human. The danger is that human judgement remains far too ready to confuse mimicry with meaning

View Full Article

Chinese universities, including those linked to the military, reportedly acquired advanced AI chips from Super Micro Computer, bypassing US export bans. This follows charges against Super Micro co-founder Yih-Shyan Liaw and others for allegedly smuggling $2.5 billion in US AI tech. The company claims it's a victim of the scheme.

View Full Article

A U.S. judge on Thursday (March 26) temporarily blocked the Pentagon's blacklisting of Anthropic, the latest turn in the Claude maker's high-stakes fight with the military over AI safety on the battlefield. - REUTERS

View Full Article

Final training runs make up just 10% of AI development costs, per Epoch AI. The rest goes to infrastructure, data, and research-before a single training run begins.

View Full Article

Boston will require all public high school graduates to demonstrate AI proficiency, making it the first major U.S. city to do so. The program launches this September, backed by a $1M grant from Kayak co-founder Paul English.

View Full Article

A growing number of U.S. college instructors are turning to oral exams to help combat an AI crisis in higher education. Some are replacing written assignments with oral exams. Others are pairing Socratic-style questioning with written assignments or requiring students...

View Full Article

Google is reportedly testing an AI tool internally called Agent Smith that can automate tasks, including coding. The tool has become so popular that the company had to limit access due to the surge in usage, reports Business Insider. Unlike traditional coding assistants, it can handle more of the workflow on its own. One of [...]

View Full Article

As Congress pushes to limit regulation of artificial intelligence systems and struggles to put guardrails on companies' handling of personal data, states continue to step up, with a key jurisdiction making moves to update its landmark AI protections and the state data privacy law patchwork expanding for the first time in nearly two years.

View Full Article

A new research reveals that reports of deceptive scheming in AI chatbots and agents have surged in the last six months.

View Full Article

People today want something different from their careers. A paycheck matters, of course, but it no longer stands alone. Workers want well-being, flexibility, and the chance to do work that feels genuinely meaningful.

View Full Article

Judge Rita Lin granted Anthropic's request for a preliminary injunction, freezing a presidential order that barred all federal agencies from using its technology.

View Full Article

Google has introduced Gemini 3.1 Flash Live, focusing on faster response times, improved voice understanding, and more reliable real-time conversations across Search, Gemini Live, and developer platforms.

View Full Article

As enterprises transition from generative AI implementations to adopting agentic AI driven automation across workflows, the security landscape transforms drastically. Attackers can no longer simply probe a model’s inference endpoint, inject prompts or leak data, they can weaponize an enterprise’s own agentic workforce. What was once a collection of agents automating procurement or finance could, if compromised, become an army of corrupt agents , issuing fraudulent invoices, authorizing false payments, or bypassing compliance workflows. As we progress to the Internet of Agents i.e. a connected ecosystem where autonomous agents across organizations, devices, and cloud environments collaborate, the attack surface also expands exponentially. Each agent becomes not just a potential target but also a potential attack vector, capable of influencing or compromising others through shared APIs, tools, or communication channels. Below are just few of the principal threat types: Memory poisoning: Adversaries corrupt an agent’s short- or long-term memory (embeddings, logs, context) to subtly steer later reasoning and decisions. Example: An attacker injects misleading entries into a customer-interaction store so future support actions and personalized offers are based on false history. Tool misuse & privilege escalation: Attackers abuse an agent’s ability to call tools or APIs to run privileged commands, exfiltrate data, or disable safeguards . This enables lateral movement across systems. Example: A compromised agent uses its automation privileges to send fraudulent emails, change access controls, and copy secure datasets to external storage. Goal manipulation: Through crafted prompts or tampering with reward/configuration parameters, attackers redefine an agent’s objectives and cause it to pursue harmful outcomes. Example: The reward function of an agent working in procurement is altered to prioritize cost reduction, which causes it to drop vetted suppliers for unverified low-cost vendors that don’t fulfill quality or compliance requirement. Communication poisoning: In conjunction with the multi-agent systems, the deceitful messages or impersonation cause disorder in the cooperation of agents and disseminate harmful instructions throughout the network. For instance, a hacker transmits misleading messages that look like they are from the Warehouse Agent, causing the Transport Agent to change the route for the high-value shipments to the fake addresses. How Secure-by-Design looks like for Agentic Systems? Secure Agentic AI by Design goes beyond traditional secure-by-design principles. Unlike other AI models, agentic systems are autonomous, stateful, and continuously learn. This means their threat surface changes over time. Security must therefore be embedded into the agent’s entire lifecycle: its memory, goals, tools, learning mechanisms and communications. The interventions can be of three types: 1. Pre-Development Agent-Specific Risk and Threat Modeling: We can apply frameworks like the OWASP Agentic Security Top 10 to identify risks and continuously scan the market for new and emerging threats. Define Boundaries & Governance into agents: Clearly articulate agent goals, permissions, oversight points, and fail-safe behaviors. Use evaluation frameworks to assess suitability of different agentic frameworks for different use-cases. Use Agentic AI management systems to track agentic ai projects and their status Data, Model Provenance & Supply-Chain Security: Vet all training and tool-sourced data pipelines to prevent poisoned inputs. Vet third-party agent frameworks and tool integrations for hidden behaviors or malicious dependencies. Evaluate the LLMs that power agentic AI against various threat vectors 2. Development, Training & Validation Sandboxed Environments: Restrict dataset and tool access within isolated training or testing contexts. Secure Memory & Tool Architecture: Use integrity-checked, versioned memory stores and enforce strong authentication for all tool APIs. A centralized, secure, and intelligent MCP registry that standardizes how Agents discover and use any internal or external tool ranging from simple scripts to production APIs can put in several layers of verification and review in tool usage. Build agent specific defenses: Purpose-built Judge and defense agents can detect malicious activity and perform assessment of agent operational intelligence, evaluating the process, not just the outcome - essential for complex tasks without a single "right" answer. Adversarial Red-Teaming: Simulate memory corruption, goal manipulation, and RCE attacks against agentic flows. Design an automated system that continuously tests different agents, models and systems against new threats and prepares vulnerability reports. 3. Post-Deployment: AgentSecOps Securing Agentic AI requires a new operational discipline: AgentSecOps is the security counterpart to AgentOps and MLSecOps. It treats every agent as a stateful entity with memory, tools, and behavioral policies that demand active defense. It aligns AI engineering, cybersecurity, and governance under a single operational umbrella. This ensures that as agents grow more autonomous, their security, accountability, and responsible alignment scale accordingly. Key principles include: Continuous Threat Detection: Monitor reasoning traces, memory states, and tool invocations for anomalies in real time using different telemetry frameworks Runtime Policy Enforcement: Implement real-time controls on tool calls (e.g., whitelisting, rate limiting, command filtering). Apply runtime guardrails that prevent goal drift and enforce least privilege for all API and code executions. Memory & Communication Integrity: Protect and monitor vector stores and inter-agent channels via encryption, authentication, and tamper-evident logging. Incident Response & Containment: Maintain rollback mechanisms, memory purging workflows, and universal “kill switch” capabilities. Human Oversight and explainability : Facilitate human-on-the-loop review for critical or autonomous decisions. Build transparent reasoning logs and decision trails for traceability. Agentic AI forms a living network of intelligence, where each agent acts as a nerve connecting systems across finance, supply chains, and data streams. This interconnected power gives enterprises adaptive intelligence but also gives attackers unprecedented reach. A single corrupt agent can ripple across the network, rewriting data or triggering automated chaos across workflows. Security, therefore, cannot be a wall, it must be the immune system of this agentic organism, embedded within every agent to detect aberrations, contain infection, and keep intent aligned. In this era of interconnected autonomy, survival depends on building self-defending intelligence, not higher barriers. By Syed Ahmed, AVP & Global Head of the Responsible AI Office, Infosys (Disclaimer: The views expressed in this article are those of the author and do not necessarily reflect the views or editorial position of Dataquest or CyberMedia.)

View Full Article

Kingfisher Chief AI Officer Mohsen Ghasempour has joined the faculty of the Global CIO Academy , which is delivered in partnership with UCL School of Management and Winmark. In a LinkedIn post, he notes that the programme brings together high potential technology leaders aspiring to become global CIOs, with a strong focus on expanding leadership impact at the enterprise level. “It is a great opportunity to contribute to this journey and engage with a highly experienced network of leaders. I am looking forward to sharing practical perspectives from building and scaling AI across retail, while also learning from the experience and insights of fellow faculty members and participants. A privilege to be involved,” he says. OK in the UK European home improvement retailer Kingfisher this week reported a rise in ​annual profit, with a strong ‌performance in the UK offsetting softer markets in France and Poland. For the year ended 31st January 2026, it booked adjusted profit before tax growth of 6% to £560 million, driven by gross margin expansion +80bps and cost discipline. The group, which owns B&Q and Screwfix ​in the UK and Castorama and Brico ​Depot in France and other markets, forecast a ⁠further rise in profit in its 2026/27 year. On the digital side of things, total e-commerce sales grew +20% excluding Screwfix (21% of group sales vs 19% FY 24/25). Marketplaces are now live across all regions. B&Q launched 15‐minute Click & Collect for 1P orders and completed the roll-out of the UK’s first marketplace Click & Collect service across 300 stores AI driven product recommendation and personalisation engines are live in all banners (excluding Iberia) - with c£165 million sales coming from this. Retail media capabilities are now available across banner apps and websites. In‐store digital media screens are in pilot at Screwfix, B&Q and Castorama France. Thierry Garnier, Chief Executive Officer, says: "We have continued to execute our strategy at pace and delivered good margin and cost discipline. This resulted in significant market share gains, profit growth of +13% when excluding last year's business rates one off and strong free cash flow.” “Our UK banners led the way, with sales +4% at B&Q and +4.5% at Screwfix. This reflects the growth of our digital ecosystem, increased share of wallet from trade customers and the opening of 34 new stores. We are making rapid progress against our strategic priorities across our banners.” He adds: “Screwfix already derives c.75% of sales from trade customers and c.60% from e-commerce. Elsewhere, trade sales increased +23% as we expanded ranges, enhanced services, and deepened relationships with trade professionals, while e‐commerce grew +20%, powered by the successful scale‐up of our marketplaces.” “E-commerce now represents one fifth of total group sales. With a mixed consumer environment across our markets, we continue to focus on delivering our strategic priorities, maintaining cost discipline and driving shareholder returns. This positions us well to capitalise on the attractive long-term structural growth opportunities within our markets." 2026 RTIH Innovation Awards AI will be a key focus area at our sister title RTIH’s 2026 Innovation Awards . The awards are now open for entries and celebrate global retail technology innovation in a fast moving omnichannel world. Our winners will be revealed at the 2026 RTIH Innovation Awards Ceremony, taking place at The HAC in Central London on Wednesday, 4th November. Check out our 2025 winners here . Our 2025 hall of fame entrants were revealed during a sold out event which took place at The HAC on 16th October and consisted of a drinks reception, three course meal, and awards ceremony presided over by award winning comedian, actress and writer Tiff Stevenson . In his welcome speech, Scott Thompson, Founder and Editor, RTIH, said: “This is the awards’ fifth year as a physical event. We started off with just 30 people at the South Place Hotel not far from here, then moved to London Bridge Hotel, then The Barbican, and last year RIBA’s HQ in the West End.” “But I’m conscious of the fact that, to quote the legend that is Taylor Swift, You’re only as hot as your last hit, baby. So, this year we’ve moved to our biggest venue yet, and also pulled in our largest number of entries to date and broken attendance records.” He added: “This year’s submissions have without doubt been our best yet. To quote one of the judges: The examples of innovative developments across both traditional and digital retail spaces were truly remarkable.” Congratulations to our winners, and a big thank you to our sponsors, judging panel , the legend that is Tiff Stevenson, and all those who attended our 2025 gathering.

View Full Article

Innovation one of 33 shortlisted from over 150 applications got under ICMR-NIN INFUSE initiative. Study protocol published preprint last April, intended product is a smartphone app.

View Full Article

Firms across the region are ramping up the use of AI, though skills gaps remain a key hurdle.

View Full Article

Critical March 2026 security updates have been released to fix multiple vulnerabilities across enterprise and AI software systems. The latest advisories highlight severe flaws that could enable attackers to execute arbitrary code, trigger denial-of-service (DoS) conditions, or escalate privileges within compromised systems. Organizations utilizing NVIDIA’s AI frameworks are strongly urged to review and patch their environments [...] The post Critical NVIDIA Vulnerabilities Enables RCE and DoS Attacks appeared first on Cyber Security News .

View Full Article

What if some of the most devastating movies from our childhood didn’t have to end the way they did?

View Full Article

Despite media narratives about "P(doom)" and robotic takeovers, a new study reveals that the general public is actually quite optimistic about artificial intelligence and its potential impact on society.

View Full Article

San Francisco, Mar 27 (AP) A federal judge has ruled in favour of artificial intelligence company Anthropic in temporarily blocking

View Full Article

Google has announced new tools for its Gemini AI that allow users to import memories and chat history from other... The post Google adds memory and chat import tools to Gemini appeared first on YugaTech | Philippines Tech News & Reviews .

View Full Article

OpenAI has put plans for a sexually-explicit chatbot on hold for good amid mounting concern about the societal and reputational risks of releasing such a product.

View Full Article

Amazon and Tech Futures Australia roll out free AI training and classroom tools for Australian schools, after research found teachers keen but under-resourced.

View Full Article

Scotland has published its first official AI guidance for schools, covering safe and ethical use for both teachers and pupils. The non-statutory guidance was developed with the EIS union and stresses that AI must support, not replace, teachers.

View Full Article