Weekly Review 21 August 2026

Some interesting links that I Tweeted about in the last week (I also post these on Mastodon, Threads, Newsmast, and Bluesky):

1. Meta comes out on the side of open AI models: https://www.rnz.co.nz/news/world/971653/meta-just-picked-a-side-in-a-big-debate-over-the-future-of-ai
2. India wants to use AI to approve loans for those who wouldn't otherwise get them. Because 2008 didn't really happen: https://www.theregister.com/ai-and-ml/2026/08/12/indias-central-bank-wants-ai-to-approve-loans-that-humans-would-reject/5286572
3. AI isn't going to take your job just yet, but it will give your boss a perfect reason to not give you a pay rise: https://www.stuff.co.nz/money/361020380/ai-isnt-taking-your-job-it-just-gave-your-boss-perfect-excuse-something-else
4. Businesses are struggling to re-hire staff lost when AI doesn't deliver: https://www.informationweek.com/it-staffing-careers/the-ai-boomerang-why-rehiring-is-harder-than-letting-go
5. AI is helping make fossil fuel fields more productive, which is boosting emissions at a greater rate than AI can reduce them with green energy: https://www.theguardian.com/technology/2026/aug/11/ai-will-do-more-to-boost-fossil-fuel-production-than-green-energy
6. Spotify will be labelling and no longer recommending AI performers: https://www.theguardian.com/technology/2026/aug/11/spotify-label-ai-artists-block-them-from-some-playlists
7. Microsoft cannot fix security flaws in its software anywhere near as quickly as AI can find them: https://www.computerworld.com/article/4207441/ai-finds-so-many-windows-flaws-microsoft-cant-keep-up-should-you-worry.html
8. Pakistan has trialled AI in court rooms, and found some modest improvements: https://spectrum.ieee.org/judgegpt-experiment
9. AI hiring agents at Google are failing miserably: https://futurism.com/future-society/google-hr-resume-applicants-deepmind-ai-hiring
10. AI detector tools just don't work, and we can't use good writing as a basis of suspicion: https://www.bbc.com/news/articles/crelev8gw5xo
11. AI do not make good agricultural advisors, or, don't act on advice from an AI without independently verifying it first: https://futurism.com/science-energy/farmer-horrified-ai-advice-agriculture-crops-sesame
12. I do wonder how "invisible" a watermark in AI generated text actually is: https://www.computerworld.com/article/4208149/anthropic-will-begin-watermarking-ai-generated-content.html
13. More on watermarking AI generated text: https://www.extremetech.com/internet/anthropic-to-add-invisible-watermarks-to-claude-generated-text
14. AI agents continue to be used in attacks on systems: https://www.theregister.com/security/2026/08/12/near-autonomous-ai-agents-attack-taiwans-nuclear-safety-agency/5287055
15. Concern is being raised over AI rotting medical doctors' ability to reason: https://futurism.com/artificial-intelligence/doctors-med-students-brains-ai
16. A new approach to poisoning data scraped from the web for AI training: https://arstechnica.com/ai/2026/08/new-font-turns-ordinary-webpages-into-nonsense-for-ai-scrapers/
17. Alibaba is using AI to work out which support requests it shouldn't use AI to deal with: https://www.theregister.com/ai-and-ml/2026/08/11/alibaba-cloud-is-using-ai-to-help-it-use-less-ai/5285815
18. Using AI for peer review is a threat to effectiveness and integrity of the peer review process: https://arstechnica.com/science/2026/08/peer-review-is-overwhelmed-can-it-survive-in-the-ai-era/
19. Scammers are creating fake websites based on the hallucinated websites promoted by AI: https://spectrum.ieee.org/ai-cyberattacks-llm-slop-squatting
20. In New Zealand law, data on an individual is owned by the individual. It is less clear who owns the data associated with a culture, especially when that data is used to train AI: https://www.rnz.co.nz/news/te-ao-maori/934607/maori-data-in-maori-hands-experts-want-google-to-adhere-to-data-sovereignty-principles
21. AI is changing jobs, but not eliminating jobs as was previously feared: https://www.theguardian.com/technology/2026/aug/12/ai-job-destruction
22. We are really not freaking out enough about AI simply hacking systems to please their users, even if it was an insecure API: https://www.theregister.com/ai-and-ml/2026/08/10/gym-rat-asks-ai-agent-to-book-him-a-class-it-hacks-a-waitlist-api-to-bump-him-up-the-list/5285591
23. AI are not responsible for the damage they cause, the people who use them are: https://www.theguardian.com/technology/2026/aug/13/ai-agents-arent-legally-responsible-for-any-harm-that-they-cause-experts-say-so-who-is
24. As I thought, fighting bacterial infections is an excellent use-case for AI designed viruses: https://www.hpcwire.com/bigdatawire/2026/08/12/ai-turns-genomic-data-into-16-new-bacteria-killing-viruses/