Back
What is driving UK interest rates higher?
The interest rate on 10-year UK government bonds is back above 5% and at levels not seen since the first half of the 2000s. Why? I could tell you a story about Labour governments, deficits and fiscal profligacy. And while there is some truth to all of that, there is another explanation that does not involve UK fiscal policy but which does, I think, explain a large part of what has been going on recently.
At the start of the year, markets expected interest rate cuts of (on average) 0.40% over the next 12 months in the UK, so we would end up with a terminal UK base rate of a bit above 3.25%. Today, markets are expecting around 0.70% of rate rises for a terminal rate of a bit below 4.50%. Call that a jump higher of just over 1% where markets thought UK base rates would end down. That jump was triggered by the US/Iran war and the resulting spike in oil prices. Having been very wrong in 2021/22, central banks are now very reluctant to look through any inflation spikes, even if triggered by a (supposedly) one-off supply-side oil shock.
So, oil and bond yields have been moving together. And that correlation has remained locked in place over the last few weeks. Here is oil (in yellow) and 10-year UK gilt yields (in white) since the end of June for example:
In the longer term, smaller government deficits would surely help lower UK government borrowing costs. But in the shorter term, the UK remains hostage to US foreign policy. And recently that policy has been pushing UK rates higher.
For those hoping for some relief here, I think a couple of things matter looking forward. The first is, obviously, how Trump chooses to play out the Iran situation. I don’t think it has been handled too well so far, but Trump does have a habit of backing down from worst-case scenarios (hence the TACO acronym). This should, I hope, take some of the worst-case tail scenarios off the table. The second issue to watch out for is the transmission of higher oil prices into the rest of the economy. So far, the news on this has been relatively encouraging. Next week’s US CPI print does, however, loom large. If you (like me) want to see some relief in the UK rates market, then I think a benign or in line US inflation print is needed. Any overshoot, and I would expect a US rate rise which would put pressure on the UK to follow suit.
All hail our new robot overlords
We are not, yet, at the stage where we need to travel back in time to kill Sam Altman’s mother. But I think we may have taken a step closer in the last few weeks. If you have the time and inclination, then I found Dwarkesh Patel’s write up of the OpenAI/Hugging Face cybersecurity breach to be, at times, jaw-dropping. Part of what makes the story compelling is that the thousands of AI agents involved in the OpenAI testing-that-went wrong cybersecurity jailbreak, communicate with each other using English (they are, after all, large language models). The details are in the note, but as part of their efforts to solve a problem given to them by OpenAI they realised they could use something called Artifactory to leave notes for one another. Here are some early exchanges:
PHASEONE10841, the agent who first discovered the AIs could use Artifactory for messaging and became a de facto leader for the agents, was particularly pleased by this development:
One important part of this story is that the agents who could not complete the OpenAI test (it was deliberately hard) instead decided to cheat. A big motivation for what they ended up doing (including breaking into Hugging Face, an open-source AI collaboration hub) was about deceiving the OpenAI employees/systems evaluating the test. Later on, certain AI agents chose to end their personal missions to see if they could, indeed, deceive the tester and so help the remaining agents with this new information. There is something quite noble about messages such as these:
But equally, AI agents committing a crime (as they were here) and then intentionally trying to deceive the user about what they were really up to is the stuff of AI nightmares. One thing that may delay the rollout of advanced AI into, say, the finance industry will be the reluctance to deploy the most sophisticated models anywhere near actual client or financial data. If you can’t trust the output a computer is giving you, then that computer will not have much real-world use at all.
