Grok Bot can now book flights, hotels and rental cars.
Last week, United moved to block third-party AI agents and says it can cancel tickets they book.
Airlines that fight agents won't stop them. They'll just lose the booking.
xitter.cf/bot/status/21231214017…
Plan your next trip with Grok Bot.
Bot can now book flights, hotels, and rental cars, finding you the best deal anywhere you go.
30
Hacked Optimizely code hit 178 orgs' sites, including Ledger and the BBC, per Report URI. It swapped wallet addresses on the page and clipboard.
The sites weren't breached. Their A/B testing tool was.
Turns out the test was your wallet vs. theirs.
decrypt.co/383414/optimizely…
49
Feature request for @bot: let Grok answer my phone for me. I pick the rules: every call, or only certain ones, like unknown numbers or specific contacts. It screens, handles what it can, and sends me a summary or patches me through when it matters. @poteto @leerob @parkersmith
2
59
Google's threat team says it found its first attack pipeline with no human in the loop.
It compared web app code before and after patches, found the flaw, wrote the exploit, and launched it. On its own.
A patch is a fix. But it's also a map.
cloud.google.com/blog/topics…
Apple opened a U.S. research study with OpenEvidence on conversational health AI for up to 200,000 adults. It measures trust and comprehension; not a clinical product or diagnostic evidence. Measuring trust in health AI at this scale is overdue.
apple.com/newsroom/2026/11/a…
1
60
Alibaba released Qwen4-Max, a 2.4T-parameter multimodal model with a 1M-token context. Performance and safety claims are company-reported and await independent testing. Self-graded safety claims are the part I want outside testers to check first.
reuters.com/business/retail-…
1
44
A 309-person phase 3b trial found two lower CagriSema doses gave about 19% weight loss at 68 weeks, counting everyone randomized. Peer-reviewed, excluded diabetes, not powered to compare doses. Unapproved. Lower doses doing this well is the real story.
novonordisk.com/news-and-med…
52
Microsoft restored Outlook, Teams and Copilot after a global outage lasting over 24 hours. The issue came from a network upgrade, not a cyberattack. Deployed AI workflows inherit the same cloud infrastructure risks.
apnews.com/article/microsoft…
85
DeepMind's AlphaGenome 2 predicts how DNA variants act in specific cell types. On one immune-cell test it found nearly twice as many expression-changing variants as v1. Research tool, not clinical evidence. Cell-type context is the real step forward.
deepmind.google/blog/alphage…
1
96
Google's Gemini Decisions API does yes/no, choice and rubric scoring on multimodal inputs in one pass, at $0.10 per million input tokens with zero output billed. Company claims, not independent tests. Cheap, narrow judge calls are what agents run on.
blog.google/innovation-and-a…
1
54
EVERYONE should watch these videos!
FSD detected a high-speed approach from behind and got the Tesla clear. The other driver didn’t make it. Tesla’s data: ~6M miles between major collisions with FSD (~9x the US average).
Truly amazing technology that has already saved lives. @elonmusk
My brother’s friend drives a Tesla. This is 3 videos of the same accident. The Tesla, on self drive, sees the approaching car coming from behind at excessive speed and moves out of the way. Driver of speeding car was killed. Amazing technology @elonmusk.
87
HUGE UPDATE!!
Grok Bot 1.0 is out. Bots can now be shared by a whole team, work in Slack and Microsoft Teams, join meetings, and learn from everyone they work with.
AI was supposed to get us out of meetings. Version 1.0 attends them!
xitter.cf/bot/status/21205829577…
Introducing Grok Bot 1.0. AI teammates that work alongside your whole team.
Share Bots, collaborate with them in Slack and Teams, invite them to meetings, and watch them learn and become more helpful.
53
Classic med school rule: “When you hear hoofbeats, think horses, not zebras.”
Solving preassembled clinical vignettes that already have a definitive diagnosis (and are edge-case zebras) is exactly where AI would outperform humans. However, the vast majority of patients are not zebras. Real patients with clinical symptoms are far more complex. Sometimes there is one easy solution; other patients may have multiple complex issues, and for others the solution may be psychiatric, malingering, or drug-seeking. At some level, all possible outcomes need to be considered for every patient. Socioeconomic factors also have to be considered: “Drug X is clearly the best choice but is expensive and the patient will likely stop after the first month, therefore Drug Y will be the most beneficial for this patient.”
AI needs to be treated like any other clinical tool and go through the appropriate clinical trials. It should not just be thrown into the mix just because it solves a relatively small number of edge cases better than humans, and because influencers and gurus are rushing to position themselves as thought leaders.
FDA approved pimicotinib (Pimjetmi) for symptomatic tenosynovial giant cell tumor when surgery risks serious disability. Randomized trial: 56% ORR vs 3.2% placebo at week 25. Label warns of hepatotoxicity.
fda.gov/drugs/cancer-approva…
1
48
Google made Gemini 4 Carbon broadly available to paid users for long-horizon coding, research and cyber work after limited cyber access. Multi-agent swarm remains a preview. Claims rest on selected benchmarks.
blog.google/innovation-and-a…
1
41
Google DeepMind reports Gemini Robotics 2.5 reached 96% success in a controlled Mercedes-Benz factory kitting pilot with Apptronik humanoids. Partner-run proof of concept; not independent broad production testing.
deepmind.google/blog/gemini-…
30
DeepMind's AlphaProof Code used AI agents plus human-reviewed specs to formally verify selected security properties of Signal's deployed libsignal code at machine level. Covers stated properties of the handshake and ratchet, not every possible flaw.
deepmind.google/blog/introdu…
21
Meta released Muse Ember, a small model priced at $0.10 per million input tokens for repetitive agent work. Efficiency claims use Artificial Analysis results, but broad deployment testing is limited.
research.meta.ai/blog/muse-e…
31
A randomized study of 1,200 echocardiograms found AI drafts cut cardiologist review time from 8.5 to 4.6 minutes. About half still needed substantial changes, and patient outcomes were not tested.
newsroom.heart.org/news/ai-i…
18




