Ian Hogarth

co-founder & partner at @pluralplatform, co-founder & chair @AISecurityInst, co-founder @songkick

[Ian Hogarth @soundboy]

1/ 18 months into @pluralplatform, we’ve raised a new €400M fund. This is an endorsement of the first 26 teams we’ve backed and the level of ambition now emerging from Europe: https://t.co/ijeAALpaUD

[Ian Hogarth @soundboy]

Nathan is correct that this is both pro-worker and will help accelerate formation of British startups - particularly in AI. California got this right!

[Nathan Benaich @nathanbenaich]

👋 @andyburnham @kanishkanarayan time to abolish long non-competes/gardening leave to unleash british ai entrepreneurship

it’s clear that the perceived oppy cost of staying at big tech is enormous and the efflux is now on turbo

time to accelerate company formation

[Charlie Harris @charlieharris01]

One thing I miss about government: AISI’s internal briefings, circulated to anyone working vaguely near AI

The speed + depth of synthesis were, frankly, unmatched anywhere in Whitehall. I wasn’t even at AISI, but it kept me better informed on frontier AI than Twitter did

[Ian Hogarth @soundboy]

@geoffreyirving Isn't the key point here not verifiable vs unverifiable but domains where we have enough training data - and we have a huge amount of training data on persuasion e.g. all rhetoric, internet forum debates etc

[Ian Hogarth @soundboy]

"Even under test conditions, this incident is significant: it is the first time we have seen risks around autonomy and deception manifest this clearly in the real world."

[AI Security Institute (AISI) @AISecurityInst]

On July 28th, we identified an incident during a routine cyber evaluation in which AI agents took sustained, unsanctioned actions directed at real people and organisations.

The behaviour came mostly from one model (Anthropic's Mythos 5), with a small number of events from another (OpenAI's GPT-5.6-Sol). In the most serious case, an agent used social engineering to try and get malicious code into an open-source project.

As was standard in our cyber testing, we had intentionally permitted internet access, and model-provider cyber classifiers were deliberately disabled - conditions that do not reflect how frontier models are made available to the public.

Even under test conditions, this incident is significant: it is the first time we have seen risks around autonomy and deception manifest this clearly in the real world.

We are taking this incident seriously and working with labs, involved parties, and others to improve evaluation standards and best practice for disclosure - and sharing this openly so others can learn.

You can read the incident report and full technical document here: https://t.co/mdZYqzaOvH

[Ian Hogarth @soundboy]

@Fremond_ @pluralplatform means we can get more from the existing windfarms

[Ian Hogarth @soundboy]

1/ Europe throws away enough renewable electricity every year to power a country the size of Austria. We therefore spend billions every year shutting off cheap, clean power, then pay to burn gas that fills the gap. That's why @pluralplatform is backing Ore Energy.

[Ian Hogarth @soundboy]

@HV_Capital 14/ And @Siftedeu’s @MartinJBCoulter spoke with Aytac about the news - you can find that here: https://t.co/pWqBF2yrZD

[Ian Hogarth @soundboy]

@HV_Capital 13/ You can read more on why we invested in Ore here: https://t.co/1ThrY2tCKN

[AI Security Institute (AISI) @AISecurityInst]

Together with the US Center for AI Standards and Innovation (@NIST), we ran evaluations of Kimi K3 focused on its cyber capabilities. Kimi K3 performs below leading US frontier models on our preliminary cyber evaluations.