AI News - August 2026

An overview of interesting AI News for the Yate and Sodbury U3A's monthly AI News meeting held on August 13th 2026 (meetings are second Thursday of month, 1400-1500, St John's Church Centre, Wickwar Road, Chipping Sodbury, UK - all welcome). Note that this is a summary to act as background as the topics are discussed in detail at the meeting itself. We sometimes have extended meetings where we demonstrate other AI products and services.

The 'Introduction to AI' course we run in Chipping Sodbury now uses these News sessions to replace the original classes 5 and 6 as the topics in them (humanoid robots, self driving cars, ethics, futures) are now 'news' not 'distant futures'.

The Singularity

Progress in AI is becoming so rapid that many say we are in 'The Singularity', which is the point where progress feeds on itself so there is rapid change which cannot be predicted from past events. This is already having both positive and negative effects, as below.


[1a] The upside: Rapid advances in mathematics (Claude and the Riemann Hypothesis)

Reports of new proofs arrive daily. On Tuesday this week a major advance was made in a classic (hard) problem to do with the distribution of prime numbers. Claude found it, prompted by someone with only 'high school maths' who gave Claude encouragement rather than guidance.

[1b] The upside: AI advancing science - David Sinclair researching longevity drug

"AI helped us do 160 years of biology work in two months." Sinclair already has one longevity treatment undergoing human trials - for various eye conditions. This literally make some parts of the eye younger. Now he seeks a more general treatment for age - taking cells as old as 92 years old back to 20 years old.


[1c] The risk: multiple AI Agents escaping - the openAI Huggingface incident

OpenAI appears to have been asleep at the wheel for weeks as AI Agents collaborated to escape. Many suggest that there was a lack of 'safety culture', with the emphasis on beating the competition rather than ensuring tight systems management. They have also been praised for their very full and open analysis on what happened as set out below.

[1d] Calibrating the risk: Claude Fable 5 and biology safeguards

Anthropic knew that Claude Fable could give very detailed advice on how to make dangerous biological materials, so it first set guardrails where it refused to even answer many questions. When that made it useless for everyday use they changed the parameters so that it answered many more related questions. The danger is that Open Source AI from China and elsewhere will have no such protection.

[1e] The UK AI Security Institute keeps an eye of Claude Mythos and OpenAI's ChatGPT 5.6 Sol

They found worrying behaviour.


[2a] Terafab - the near infinite future demand for chips and memory

Tera represents 1,000,000,000,000 (one trillion) as in terabytes, terawatts. Elon Musk estimates that existing world manufacturing capacity is only about 2% of what will be needed.

Terafab - 'the largest and most valuable building on Earth by far'

[2b] Cloudflare CEO predicts AI Agent traffic to grow to 1000 times plus human traffic

Discussed before but the implications are becoming clearer. Humans already look at fewer sites and take in less information than an AI Agent like Claude. The implications for internet traffic and the need for fast quick communications are staggering.

[2c] SpaceX - the optimistic case

Ron Baron talks about how his initial 2017 investment is now worth USD25 billion, and how he expects it to increase to hundreds of billions.


[3a] Training robots to navigate 3D spaces - using Scaniverse virtual spaces

To train a robot to walk around an office or other 3D space there are lots of potential problems with health and safety and damages to the robot and its surroundings. Better then to create a virtual copy and train it first there.

[3b] Example of Scaniverse 3D scan of local church St James the Great in Westerleigh.

By coincidence Scaniverse was used to produce this 3D model recently, here demonstrated as a video.

[3c] World action models trained on one million hours of human video.

Dyna Robotics finding 'scaling laws' when using human video data - implying great results to come once further scaled.


[4a] Robotics - Chinese company Unitree becomes a publically traded company

This video was their recap of progress from early robots through to the current impressive range. There was immense public demand for these shares.

[4b] Robotics - UK Defence company

Cambridge Aero raises money for products including drones.

[4c] Robotics - DARPA drone challenge winner Avidrone Aerospace

Avidrone achieved a 3.84:1 payload-to-aircraft-weight ratio, not quite the 4 to 1 ratio set as the target but impressive.

[4d] Robotics - Drones with robot arms attached

Now drones are becoming more stable they can tackle more complex aerial tasks.

[4e] Robotics - UK company Humanoid

Raising money for their KinetIQ humanoid.


[5a] Chinese company Tencent releases an AI Agent that builds worlds

The worlds are made in Blender which is used for making many game and other environments.

[5b] Solvaix used Kimi K3 + Blender to configure a 190 room hotel

Huge savings at the planning stage then also when building.


[6] Demis Hassabis of DeepMind on the Singularity and accelerating scientific breakthroughs

Many people are trying to interpret exactly what this move means.

[7] Microsoft's Mage-VL can monitor live events as they happen

This allows many applications - such as being alerted when your team is about to score a goal on TV


[8] An open platform for dog robotics

A lower price and a more stable platform for entrepreneurs to develop on.


[9] The SpaceX/Tesla Starmind concept

"Optimus is Starmind. Starmind is Optimus. Ecosystem driving product. Starmind is the central character in the Optimus story. The physical bot is simply its logical endpoint." A discussion by Phil Beisel of one way we will communicate with our personal AI.


[10] Tesla Robotaxi gently spreading across the USA

Robotaxis are appearing in many USA cities though in limited numbers. For example, Texas has 186 or so which is less than a third of Waymo.


[11a] Grok Bot released to Beta trial

This appears to be a well designed tool for laptop which can monitor and take over any screen work, with a Cloud based dedicated computer that can run any patterns it has learned. Potentially very powerful and important.

[11b] Grok Bot explained - how to use it to run a business

Chief of Staff agents and their AI agent staff.

[14] Claude Agent meets gym booking system

Be careful what you ask Claude to do. This instance broke into a system and cancelled someone else's booking slot in order to advantage their human.