AI for the rest of us

Welcome to the AI community for everyone.

Hello friends,

Hannah is home from Wales and in a very “back to school” kind of mood. It’s a great feeling to be excited about going back to work! This week has been busy in the best possible way with dates for AI for the rest of us 2027 now confirmed.

Save The Date: 25-26 February 2027

Inspiring keynotes, four tracks and a workshop room for hands-on learning. Super early bird tickets go on sale in this newsletter next week!

Charles has been crunching away at work, but he did find time to read Raising Hare by Chloe Dalton, who during the pandemic retreated from London to a restored stone barn situated somewhere in the English countryside, distancing herself from her work in the city as a well-connected political adviser (the book’s acknowledgments feature William Hague and Angelina Jolie). She reluctantly rescues a leveret after finding the creature lying on a country track outside her home, seemingly abandoned and, over the course of the book they develop a remarkable relationship. It's rather wonderful.

Have a fabulous week!

Hannah and Charles

Did AI Break Recruitment?

Join us for the next London meetup on Tuesday 29th September.

What’s Hannah reading this week?

There’s a lot to share after taking a week off. Some of the news probably isn’t even news anymore but hopefully it’s still worth reading about.

With SpaceX’s acquisition of trailblazing AI Coding platform Cursor, OpenAI announced that it was cutting off Cursor’s access to their models. It’s not surprising given Cursor competes directly with Codex, OpenAI’s coding tool.

Earlier this year Cursor was our chosen coding platform. You had access to all the frontier models alongside Cursor’s own cheaper models, which were great for less demanding work. However, Cursor gave us a glimpse of how good OpenAI’s models had become and we got hooked. Cursor’s strict limitations on premium models became a business problem and we switched to Codex.

(You might be wondering why we didn’t switch to Anthropic. Quite simply, it was the price. We blew through our token limit too quickly leaving us stuck between downing tools or spending more. Our budget is in the hundreds not the thousands...)

This move is obviously a blow to Cursor users but it could also cut off a sales pipeline for OpenAI. I wonder how many people, like us, used Cursor to assess the performance of different models. Inside Cursor OpenAI were able to passively showcase their models directly alongside others without any effort. Now they can’t. It will be interesting to see how many people stick with Cursor or switch.

In the world of cyber security I recently discovered abliteration.ai - this team is proudly removing safety guardrails from open models (most recently GLM 5.3) and redistributing them on their platform. Models engineered to refuse cyber attacks can now be used for red teaming (i.e. ethical hacking). This is good news for defence and attackers alike and I’ll be interested to find out more about how abliteration intends to tell the difference.

In a recent blog Fin Tech startup Seapoint have shared why they built an internal agent platform. The story of Harbor is a little light on technical details but I found it interesting that this team had gone “all in” on empowering business users to build their own agents. Unlike the story I shared a few weeks ago from Uber whereby they paired engineers with business users, Seapoint have decided that engineers should focus on the platform instead of the agents.

"At first, we tried an open-source agent platform, but it was built for developers, with no realistic way for a non-technical person to build an agent through the UI."

This story also reveals a huge gap in the AI ecosystem, a platform that effectively abstracts all the tech away from a business user so they can build what they need. I want this platform to exist! I want to use it myself!

The cyber resilience act was first announced in 2022 and this week finally came into effect. You were given plenty of time, I hope you used it wisely! If you didn’t and you’re wondering what it means to you this excellent guide from OpenSSF might help. Mandatory vulnerability and incident reporting requirements apply from 11 September 2026 so there’s nowhere to hide.

At my startup BIMP we want to help teams eliminate vulnerabilities in their containers. There are plenty of zero CVE options available to choose from, but for a lot of teams finding and replacing all your current base image configuration is messy and manual. Yes this is a shameless plug, but if you’re worried about CRA revealing the truth about your vulnerability counts now might be the moment to switch to a zero-CVE alternative. Let me know if we can help you with that!

What's Charles reading this week?

As mentioned I’ve had a busy week, but I did see that Apple unveiled its latest iPhones, watches, AirPods etc. Their first foldable iPhone Duo looks stunning (and expensive) but I think Siri Recap might be an equally big deal. It looks like you can set it to listen all day, taking notes but not keeping audio or a transcript, and all private on your watch. There are all sorts of privacy issues with this, but it is also the kind of ambient intelligence that could be really useful.

You will almost certainly have seen that Evan Hubinger, a safety researcher at Anthropic has warned that AI is advancing so quickly he believes there is a greater than 10% chance it "could kill all humans" within the next decade. He posted this after another Anthropic researcher Jacob Coxon publicly announced his resignation on X late Monday over the same broad concerns.

What are we to make of this?

An AI that can continually improve itself, potentially to a point beyond human control or understanding, has been a long-standing theme of both the dystopian science fiction which forms at least part of Gen AI’s training data and also parts of the AI research community. There are good reasons to be concerned about the impact of AI, and Hubinger and Coxon obviously have deep knowledge about the technology. But I think the whole thing is a distraction.

If you can make everyone believe that it's not negligence that’s the problem, it's “the models going rogue”, then maybe you won’t be held accountable for your carelessness. Likewise, getting people worried about near future existential risks, takes the focus away from the very real impacts of AI’s excessive energy use. After all, why worry about that if we’ll all be dead in 10 years anyway?

So I found myself in extremely strong agreement with every point Thomas Claburn made in this opinion piece for The Register:

“Coxon said, "These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources. … No other human activity poses this level of danger."

Here's one: Human-induced climate change. In 2023, according to researchers, more than 178,000 deaths can be attributed to a global heat wave. "More than half (54.29 percent) of heatwave-related deaths were attributable to human-induced climate change," they claim.

That's 96,636 deaths attributable to human activity – or perhaps lack of it – just in the context of a heat wave.”

Claburn goes on to argue that society could pass laws to put executives in jail when their models do harm which, honestly, isn’t a bad idea.

With all this going on Anthropic has also revealed that it has already disrupted several plots by scientists this year that could have helped to develop biological weapons, as part of a larger "threat intelligence report" detailing efforts to use Claude for surveillance, cyberattacks, influence operations, and more.