Welcome to another Bi-Weekly Frequency. Through here, you’ll be able to access a roundup on all things B2B tech, algorithmic changes on the most popular communication channels, and top news of the week, analysed by members of our team.
Last time, we followed LinkedIn’s rise as a source for AI search and looked at the tougher safety bar for frontier models. This week, the feeds are handing control back to their users, while in Westminster, MPs and MI5 are asking who gets to control AI itself.
Algorithmic changes: the feed gets a dial
For years, the feed decided what you saw. This fortnight, two platforms started letting people turn the dial themselves.
On Wednesday, Substack launched new content preferences. Users can now see more or less video, audio, long-form or community content, mute specific keywords and phrases, fine-tune the topics they follow and automatically label AI-generated material. Substack has a reason to invest in its feed: over the past six months, the app drove nearly 800,000 paid subscriptions and more than 80 million free ones, with more than half of the paid subscriptions coming straight from the Notes feed.
LinkedIn is moving in a similar direction. Chief Product Officer Hari Srinivasan announced on 18 September a test of a “Network” tab, sitting next to the “For You” feed on mobile. Network shows only posts from people a member is connected to or follows, while For You stays focused on discovery, from inside and outside someone’s network. It’s still unclear whether company pages will appear in the Network tab.
A caveat, though: this is a test, and LinkedIn has run a few before. Last year it experimented with several alternative feeds, and in June it tried topic-based suggested feeds, but neither got a wider launch. As long as For You remains the default, the impact should stay limited.
Our takeaway: when readers can mute, filter and label, the algorithm stops doing the work for you. That leaves two routes. Earn a place in the network feed, which means people following people, so executive and employee voices will matter more than the company page. And give the discovery feeds something worth surfacing. A specific, substantial post survives a keyword mute. A generic trend take that reads like AI is exactly what these new controls are built to filter out.
News of the week: one day, three warnings
On 13 October, the IMF (International Monetary Fund), Parliament and MI5 all raised the alarm on AI within the same 24 hours.
At its annual meetings in Bangkok, the IMF nudged its global growth forecast up to 3.1%, largely thanks to AI investment offsetting higher oil prices. But its Global Financial Stability Report named a potential AI share crash as one of three key risks to financial stability, alongside rising borrowing costs and a major cyberattack. Managing Director Kristalina Georgieva summed it up: the economy is “standing on one leg, the AI leg.” For B2B tech, that means buyers will start asking harder questions about return on investment, and “AI-powered” will no longer close a deal.
In Westminster, the Business, Innovation, Science and Trade Committee opened a long-term inquiry into AI safety and governance. Former OpenAI and Anthropic staff who left over safety concerns gave evidence first, and their warnings were then put to OpenAI, Anthropic, Meta and Google DeepMind. OpenAI pointed to the cancelled GPT-6.1 Astra as proof it isn’t racing to the bottom. Chair Liam Byrne wasn’t convinced, noting that officials admitted the Government still hasn’t decided how or when it will act. The inquiry will stay open, with targeted calls for evidence on specific issues.
The same day, MI5’s Sir Ken McCallum announced a new AI Threat Assessment Centre, working with the AI Security Institute, the National Cyber Security Centre and the National Crime Agency. He said terrorists had already been caught using AI to improve their attack methods.
It sounds like three separate warnings, but the message is the same: the burden of proof is moving onto the companies. Last time, a voluntary accord was followed by a regulator within a day. This time, the economy, Parliament and the security services are all checking. It is likely that calls for evidence and investor scrutiny will become routine for anyone selling AI.
The bottom line: