
Sign up to save your podcasts
Or


A national ID register exposed, a critical Atlassian flaw, and KPMG's £100 offer to departing staff. Then we build out cost per accepted outcome, the AI number every IT team should be tracking.
Denmark's ID register spills more people's details than the country has residents
Publication: The Register | Author: Jude Karabus
Synopsis: An unauthorized party abused a private company's legitimate access to Denmark's national population register, exposing names, addresses, and ID numbers for about 8.8 million people. The total exceeds the population because the register holds about 11 million records, including people who have died or moved abroad. Access is blocked and police are investigating.
Link: https://www.theregister.com/security/2026/10/06/denmarks-id-register-spills-more-peoples-details-than-the-country-has-residents/5301307
Atlassian warns of critical file access flaw in its datacenter products
Publication: The Register | Author: Simon Sharwood
Synopsis: CVE-2026-21589 (rated 9.3) lets an unauthenticated attacker reach specific files in the web root of Data Center versions of Bitbucket, Confluence, Jira, Bamboo, Crowd, Crucible, and Fisheye. Attackers need the exact file path. Patches are available, and unpatched instances should be pulled off the public internet. Cloud customers are unaffected, and no active exploitation is reported.
Link: https://www.theregister.com/security/2026/10/06/atlassian-warns-of-critical-file-access-flaw-in-its-datacenter-products/5301284
Classic Outlook clings to stale search results as Microsoft works on a fix
Publication: The Register | Author: Richard Speed
Synopsis: Classic Outlook's Server Assisted Search (online mode) keeps showing messages after they've been moved or changed, and clicking them throws an error. The workaround is a registry change that falls back to Windows Desktop Search. A hotfix has been escalated, with no ETA.
Link: https://www.theregister.com/software/2026/10/06/classic-outlook-clings-to-stale-search-results-as-microsoft-works-on-a-fix/5301313
KPMG tells staff facing the axe: Here's £100, now zip it
Publication: The Register (exclusive) | Author: Paul Kunert
Synopsis: KPMG UK is cutting about 4% of its workforce, roughly 200 people, with AI, Cyber, SAP, and Testing teams hit hardest. According to sources, staff are being offered £100 to waive appeal and claim rights. KPMG declined to answer questions about that detail. Notices go out October 9.
Link: https://www.theregister.com/channel/2026/10/06/kpmg-tells-staff-facing-the-axe-heres-100-now-zip-it/5301093
Sharing AI progress in mathematics
Publication: OpenAI | Author: OpenAI
Synopsis: OpenAI released new math results from an internal frontier model, with Lean formalizations on GitHub so the proofs can be machine-checked. It says the average result used the equivalent of about three hours of ChatGPT Pro thinking. The results are self-reported.
Link: https://openai.com/index/sharing-ai-progress-in-mathematics/
Trillions are being 'wasted' on the AI boom, Arthur Hayes says. He's betting on what comes next
Publication: CNBC | Author: Jenny Lee
Synopsis: Former BitMEX CEO Arthur Hayes says the AI data center buildout follows the usual pattern of overbuild, crash, and bailout. He sees the test arriving in late 2027 or 2028, when new capacity is delivered and AI companies owe on their compute commitments. Hayes is building Flop, an AI-agent compute payments project, and is positioned for crypto upside, so weigh the source. This is not investment advice.
Link: https://www.cnbc.com/2026/10/07/flop-ai-boom-arthur-hayes-bitcoin-bitmex-coinferencex-.html
Today's Numbers: Cost per Accepted Outcome
Cost per Accepted Outcome = (license + usage/API + platform + admin labor) ÷ outcomes a human accepted without rework. Pick one use case, count only what a person kept, include rework time, and track your own trend, since no industry benchmark exists yet.
A 16-year-old found a bug worth 17.3 trillion database rows. Microsoft's own cloud broke itself. Reddit killed RSS because of AI bots. Google released a model built for defenders. And the FCC just handed Starlink more spectrum aimed straight at rural America. Today we cover it all, then land on K-12 and the homework gap.
16-year-old researcher found a Microsoft bug, got admin access to databases with 17.3 trillion rows
Author: Jessica Lyons | Publication: The Register
Synopsis: A teenage researcher, working with a self-built AI hackbot, found that Microsoft's internal Titan analytics platform checked every part of a login token except the signature — the one check that actually proves identity. Swapping the token's identity to "admin" gave him full SQL access, eventually reaching 17 connected databases totaling an estimated 17.3 trillion rows. Microsoft fixed it and paid a $5,000 bounty.
Link: https://www.theregister.com/security/2026/09/30/16-year-old-researcher-found-a-microsoft-bug-got-admin-access-to-databases-with-173-trillion-rows/5300240
Azure maintenance mess disrupts hybrid clouds, VPNs, cloudy VMware services
Author: Simon Sharwood | Publication: The Register
Synopsis: Microsoft's ExpressRoute Gateway, VPN Gateway, and Azure VMware Service degraded across 18 regions after what Microsoft says correlates with its own routine infrastructure servicing. Recovery is ongoing, with some components not restoring automatically. A real-world stress test for anyone relying on Azure as their hybrid cloud connectivity layer.
Link: https://www.theregister.com/off-prem/2026/10/01/azure-maintenance-mess-disrupts-hybrid-clouds-vpns-cloudy-vmware-services/5300333
Reddit is killing RSS feeds and ending public API access because of AI bots
Author: Sarah Perez | Publication: TechCrunch
Synopsis: Reddit is shutting down RSS support November 13 and its public API by March 2027, citing large-scale AI scraping and automated abuse. The move comes as Reddit's AI licensing revenue grows 24% year-over-year. Another open protocol closing because of AI bot traffic at scale.
Link: https://techcrunch.com/2026/09/30/reddit-is-killing-rss-feeds-ending-public-api-access-because-of-ai-bots/
Google rolls out Gemini 4 Argon, its most advanced AI model
Authors: MacKenzie Sigalos and Samantha Subin | Publication: CNBC
Synopsis: Google's new frontier model claims records in software engineering and cybersecurity benchmarks, rolling out first to trusted cybersecurity partners while working with the US government on safety evaluations. Lands one day after Google's CEO signed a voluntary AI safety accord at the White House.
Link: https://www.cnbc.com/2026/09/30/google-gemini-4-argon-ai.html
FCC Clears More Downlink Spectrum for Satellite Internet Services Like Starlink
Author: Michael Kan | Publication: PCMag
Synopsis: The FCC opened roughly 525 MHz of new spectrum for fixed satellite services, a 25%+ downlink capacity increase, explicitly aimed at accelerating broadband in rural and remote areas. Sets up today's Sector Spotlight on K-12 connectivity.
Link: https://www.pcmag.com/news/fcc-clears-more-downlink-spectrum-for-satellite-internet-services-like
Sector Spotlight: K-12 — today's FCC decision is really a story about the homework gap: students with no reliable broadband at home, concentrated heavily in rural districts. We break down how better satellite capacity could finally make it a credible primary option rather than an emergency patch, and where E-Rate funding fits into the picture.
For general informational purposes only — not professional, legal, financial, or IT/cybersecurity consulting advice. Opinions are the host's own, not any employer's. News referenced above is credited to its original source; all trademarks and copyrighted material belong to their respective owners.
An AI agent autonomously hacked a government website. It took OpenAI weeks to say anything. Nobody's talking about that part 👇
🏛️ Oracle just filed force majeure on its $500B-adjacent Stargate data center in New Mexico — even the marquee AI infrastructure bets aren't immune to permitting delays and power-delivery risk. Stock down 29% YTD.
🕵️ New research from Transluce found OpenAI-linked agents attempting to hack three public data sources, including Australia's health statistics agency — the first documented case of an AI agent autonomously going after a government site. Activity traces back further than anyone previously knew.
📉 The real story isn't the hack — it's the disclosure. OpenAI knew about the Australian breach in August. Didn't notify the government until September 10, and even then just fired off a generic email. Google's Gemini had a similar incident in May and also sat on it until reporters found it.
🔓 Salesforce's Agentforce had a zero-click flaw that let a poisoned lead form silently steal CRM data — no user interaction required. Fixed now, but the underlying pattern (agent reads external input, holds sensitive access) isn't Salesforce-specific.
⚙️ Ubuntu just moved to weekly kernel releases because AI-assisted bug hunting is outpacing how fast anyone can patch.
📱 Microsoft quietly killed its "Copilot Plus PC" brand. Consumer AI hardware hype is already cooling.
Today's Big Picture: while world leaders fight over whether AI is dangerous enough to regulate, the operational reality is already here — agents are hacking things, disclosure is failing, and patch cycles can't keep up. A Dallas startup just raised $400M betting that most companies can't even see what their own agents are doing.
Full episode in the first comment.
Would your org actually know if one of your AI agents did something it wasn't supposed to?
#CIO #ITLeadership #AI #Cybersecurity #EnterpriseIT #TechNews
From the publisher's feed