Three of this week's four stories come back to a question I hear from almost every leader we work with right now: what are we getting for this AI spend, and what new risk came with it? Here's what I'd want you to know, and what I'd do about each. If you want a second set of eyes on any of these, talk with the Spej team.
AI can now turn a published patch into a working attack in hours.
On October 27 Anthropic published research on an internal model finding critical vulnerabilities and writing attack chains that were reported and fixed. The next day Google introduced Gemini 4 Carbon for a group of defenders first, and on October 29 Epoch AI showed a widely available model turning Firefox fixes that were published but not yet shipped into working attacks against six releases, at about $1,000 each, the fastest in 12 hours. Ask IT or your MSP how many days a critical patch takes to reach laptops, servers, and network gear.
Microsoft offered voluntary buyouts to about a quarter of its workforce.
On October 26 Microsoft offered a voluntary exit to employees at level 70 and below in the U.S., Canada, India, and the U.K. CNBC reported about 60,000 people are eligible, roughly 27% of staff, with the offer framed against the cost of Microsoft's AI build-out. Microsoft told GeekWire the offer is not a sign it needs fewer employees and is not a replacement for layoffs, so if you have a plan for what AI means for roles, tell people, and if you do not have one yet, say so and give them a sanctioned place to try tools.
Amazon told its engineers that token counts are a flawed way to measure AI.
On October 27 Amazon posted an engineering guideline that calls token usage, AI tool adoption, and code-change counts deeply flawed measures, and says tokenmaxxing is at odds with frugality. CNBC reported Amazon shut down an internal leaderboard in May after employees gamed it. Pick one workflow, write down a before number such as hours, error rate, or cycle time, and keep the usage dashboard out of performance reviews.
Anthropic's IPO filing shows how much AI gets sold through partners.
Anthropic filed its S-1 on October 26. Bloomberg reported that 48% of its revenue in the first half of 2026 came through the Amazon, Google, and Microsoft cloud marketplaces, up from 11% in 2023, and CRN reported that fees paid to channel partners rose $2.7 billion. For most midmarket companies, AI will arrive through a contract you already have, so ask who the subprocessors are, where the data goes, and what a connected agent can reach.
First published on Spej's Substack.
