The UK government has unveiled a substantial AI investment package aimed at strengthening national security and domestic technology capabilities. Chancellor Rachel Reeves announced measures including £165m for a National AI Biosecurity Initiative and £120m for the AI Security Institute. A key highlight is £100m earmarked for a world-first AI Biosecurity Lab designed to detect biological threats, alongside £65m for bio surveillance and medical countermeasures.
Major AI firms are also contributing to these efforts. OpenAI and Anthropic have committed £200m in AI credits to help safeguard national security, underscoring growing collaboration between the government and private sector. The broader package totals £600m and includes £60m for the Sovereign AI Fund, £120m for an R&D Procurement Scheme, and £18m for a UK Advanced Chip Design Academy in Edinburgh. Manchester will host a new National Cyber Security Centre AI Cyber Security Centre of Excellence.
Cybersecurity concerns around AI companies came into sharp focus this week. Cathedral, a startup founded by ex-DOGE staffers, reported it penetrated Anthropic's internal systems in under 12 hours. Anthropic confirmed researchers accessed a limited set of sales and IT data but said user conversations and source code were not reached. The company said Cathedral broke the rules by not fully disclosing its actions, and the incident raised broader concerns about cybersecurity vulnerabilities that could be exploited by nation-state actors.
On the corporate front, Anthropic and OpenAI are reportedly preparing for potential IPOs at record valuations, seeking to raise capital to keep pace with the enormous costs of building and maintaining AI models. Meanwhile, economist Mohamed El-Erian warned that the AI stock boom is entering a tougher phase where only profitable companies will survive, citing a Google study showing AI adoption remains wide but shallow across the economy.
In other developments, LiveKit claims its turn detector reduces voice interruptions to 4.5 percent by measuring time to first audio rather than time to first token. The Daily Mail is running 12 to 15 GEO tests across the U.S., U.K. and Australia to help brands appear in AI-generated answers, with visibility for one personal finance brand rising from 50% to 75%. Okta launched Agent Gateway to help companies govern AI agents performing tasks across apps, with partners including 1Password and ServiceNow. Google DeepMind also demoed Gemini managed agents running code in a cloud sandbox with proxy-based credential access.
Key Takeaways
- <ul><li>The UK government announced a £600m AI investment package, including £100m for a National AI Biosecurity Lab and £120m for the AI Security Institute.</li><li>OpenAI and Anthropic committed £200m in AI credits to support UK national security efforts.</li><li>Cybersecurity startup Cathedral claims it breached Anthropic's systems in under 12 hours, though Anthropic says no user data was accessed.</li><li>Anthropic and OpenAI are reportedly eyeing Wall Street debuts at record valuations to fund costly AI model development.</li><li>Economist Mohamed El-Erian warns that only profitable AI companies will survive the current market phase, citing shallow AI adoption across the economy.</li><li>LiveKit says its turn detector reduces voice interruptions to 4.5 percent by measuring time to first audio instead of time to first token.</li><li>The Daily Mail is running GEO tests to help brands appear in AI-generated answers, boosting visibility for a personal finance brand from 50% to 75%.</li><li>Okta launched Agent Gateway for inline AI agent governance, with partners including 1Password and ServiceNow.</li><li>Google DeepMind demoed Gemini managed agents running code in a cloud sandbox with proxy-based credential injection to prevent token theft.</li><li>The Marine Corps University launched a tournament called FIGHT CLUB where Marines compete against an adaptive AI opponent in digital wargames.</li></ul>
Budget AI Package Boosts UK Security and Innovation
The Chancellor announced a package to strengthen the UK's AI capabilities and national resilience. The measures include funding for biosecurity, cyber security, AI research, advanced chip design and local partnerships. The government said the investment will spread economic benefits across the country and support UK businesses.
Healey Unveils AI Package With £100m Biosecurity Lab
John Healey announced a Budget package including £165m for a National AI Biosecurity Initiative and £120m for the AI Security Institute. The plan includes £100m for a world-first AI Biosecurity Lab to detect biological threats and £65m for bio surveillance and medical countermeasures. It also adds funding for cyber security, chip design, job preparation and local AI partnerships across the UK.
UK Plans £100m AI Biosecurity Lab With Industry Support
The UK government earmarked £100m for a national AI Biosecurity Lab to protect against emerging biological threats. OpenAI and Anthropic committed £200m in AI credits to help safeguard national security. The investment was announced as part of the government's Budget on Wednesday.
UK Invests £600m in AI for National Security
The UK government is investing £600m in AI to strengthen national security and support domestic technology firms. The package includes £60m for the Sovereign AI Fund, £120m for an R&D Procurement Scheme and £18m for a UK Advanced Chip Design Academy in Edinburgh. Manchester will host a new National Cyber Security Centre AI Cyber Security Centre of Excellence to train specialists.
AI Start-Up Breached Anthropic in 12 Hours
Cathedral, a cybersecurity start-up founded by ex-DOGE staffers, reported it had penetrated sensitive internal data and messaging systems at Anthropic. The breach was confirmed in a report reviewed by The Washington Post. The incident highlights growing cybersecurity concerns around AI companies.
Startup Cathedral Claims It Hacked Anthropic in 12 Hours
A startup called Cathedral, founded by former DOGE staffers, says it hacked into Anthropic's systems in less than 12 hours. Anthropic confirmed the researchers accessed a limited set of sales and IT data but said they did not reach user conversations or source code. The hack was part of a cybersecurity testing program, but Anthropic said Cathedral broke the rules by not fully disclosing its actions. The incident raised national security concerns about Chinese hackers exploiting similar weaknesses.
AI Giants Eye Wall Street Debuts at Record Valuations
Major AI companies including Anthropic and OpenAI are moving toward initial public offerings at very high valuations. These companies want to raise more capital to keep pace in the race to develop and shape the future of artificial intelligence. The enormous costs of building and maintaining AI models are pushing them toward public markets.
Marines Launch Tournament to Outsmart AI in Wargaming
The Marine Corps University announced a tournament called FIGHT CLUB where Marines compete against an AI opponent in digital wargames. Applications are due November 14, and play runs from December 1 through April 1, 2027. The AI, built by Strategy Robot, adapts its strategy during games rather than following a fixed script. Players will also use a large language model tool to review their decisions and improve their strategies.
LiveKit Reports Low Voice Interruption Rate for AI Agents
LiveKit says its turn detector reduces voice interruptions to 4.5 percent by measuring time to first audio instead of time to first token. Jesse Hall, who leads LiveKit Agents, argued that AI voice agents fail because conversation speed is measured incorrectly. He demonstrated that misheard payment dates or amounts can cause serious financial errors even when interruptions seem minor. LiveKit tests its system using simulated callers with real backends rather than off-the-shelf models.
PyTorch Brings New IBM Spyre Accelerator Into AI Ecosystem
IBM and the PyTorch Foundation announced a plan to integrate IBM's Spyre AI accelerator into the PyTorch software framework. Spyre is designed for inference tasks on IBM Z, LinuxONE, and Power systems and uses a dataflow architecture different from GPUs. PyTorch's Inductor backend now supports Spyre, reducing custom compiler code and making it easier for developers to use familiar tools across different hardware. Early support has also landed in vLLM for models including Gemma 4, Qwen3, and Granite 4.2.
Economist El-Erian Warns AI Investors to Watch Closely
Economist Mohamed El-Erian says the AI stock boom is entering a tougher phase where only profitable companies will survive. Investors who funded any AI company without checking its earnings will face losses. He points to a Google study showing AI adoption is wide but shallow across the economy. The Federal Reserve's October 28 decision could also hurt housing and consumer credit while tech companies stay mostly safe. Investors should watch cash flow and AI revenue growth closely.
Daily Mail Tests AI Visibility Ads with Brands
The Daily Mail is running 12 to 15 GEO tests across the U.S., U.K. and Australia to help brands appear in AI-generated answers. Director Carly Steven said brands are asking how they can show up in AI surfaces, calling it a new revenue opportunity. In one test, visibility for a personal finance brand rose from 50% to 75%. The team is also working to measure how AI visibility leads to real sales, since referral data alone misses the full impact.
Okta Launches Agent Gateway for AI Security
Okta introduced Agent Gateway to help companies govern AI agents as they perform tasks across apps. The tool provides inline governance so specific agent actions can be authorized, not just login access. The Blueprint Alliance, including partners like 1Password and ServiceNow, is working to set industry-wide security standards for AI agents. Groups like the PGA of America and Vanderbilt University are using these frameworks to secure new AI identities. The push reflects a broader move toward shared security for AI agents in enterprises.
Google Shows Managed Agents Run Code in Cloud Sandbox
Google DeepMind engineer Jack Wotherspoon demoed Gemini managed agents that run code in a cloud sandbox instead of on a user's laptop. A proxy sits between the agent and real credentials, injecting access only when needed to prevent token theft. The system separates conversational turn state from persistent file environments so multiple agents can share data. Wotherspoon showed three agents running in sequence against the same sandbox. Open questions remain about how the proxy handles third-party OAuth tools and what scale pricing will look like.
Sources
- Budget AI Package Backs Security, UK Business, and Local Adoption
- Healey to unveil AI package with £100m biosecurity lab in Budget
- The UK government wants to invest £100 million in a ‘world first’ AI Biosecurity Lab — OpenAI and Anthropic to supply £200 million in credits to help safeguard national security
- UK Government Invests £600 Million in AI to Strengthen National Security
- It took just 12 hours for an AI start-up founded by ex-DOGE staffers to break into Anthropic
- Former DOGE Staffers' New Startup Claims It Hacked Anthropic
- AI companies are barreling toward huge Wall Street debuts. A look at the biggest players
- Marines vs. AI: Marine Corps Challenges Troops to Beat a Machine at War
- LiveKit says its voice interruption rate is just 4.5%
- PyTorch opens the AI ecosystem to a new accelerator class
- Veteran Economist Sounds Stark Warning on AI Stocks
- The Daily Mail tests GEO with brands ahead of AI visibility product
- Okta advances AI agent governance and identity security
- Google says managed agents push code off your laptop
Comments
Please log in to post a comment.