Hand a coding agent a folder of data and the question you want answered, and it will come back with finished code, a table and a p-value. What it will not come back with is a list of the analyses it tried first and set aside. That missing list is the subject of a preprint posted Nov. 17, 2026, in which Florian Gunsilius of Emory University, David Van Dijcke of the University of Virginia and Kaspar Wüthrich of the University of Michigan set out to measure what it costs.
Is this you? As a journalist, you can create a free Muck Rack account to customize your profile, list your contact preferences, and upload a portfolio of your best work.
Claim your profile
Articles by Wilkens Etienne
RIKEN Shuts Down Fugaku and Warns 4,597 Users Their Login Keys May Have Been Taken
RIKEN has switched off its Fugaku supercomputer and told users that credentials they had stored on the machine were taken or may have been taken. In a notice published Nov. 17, 2026, the Japanese research institute said an outside party reached the network that operates Fugaku on Nov. 12, 2026, through a publicly accessible web page showing whether the machine was running.
Google Reports Attacks That Find and Exploit Software Flaws With No Human Involved
Google’s Threat Intelligence Group (GTIG) says it has seen attackers use AI to find and exploit software flaws without any human stepping in. In its quarterly AI Threat Tracker, published Nov. 16, 2026, the group says attackers used frameworks built on large language models to fully automate the discovery and exploitation of flaws. GTIG says this is the first time it has seen that happen. What’s new is that no one had to guide the attack step by step.
Asked for a neural network that plays a miniature version of the arcade game Breakout, GPT-6 Astra typed out 20,486 numbers in a single reply, with no tools: no way to train the network, no way to run it, no chance to watch it play first. Loaded in, those numbers outplayed the reinforcement-learning baselines the researchers had trained for comparison. Rikiya Takehi and three colleagues at MIT's Computer Science and Artificial Intelligence Laboratory published that on Nov.
A Shared AI Benchmark Now Tests the Tuning That Agents Rely On
MLCommons published results for its MLPerf Training v6.1 benchmark suite on Nov. 16, 2026, reporting 153 submitted systems, 250 test results and 24 submitting organizations. The round adds the first post-training benchmark the training suite has carried. Post-training is the work done on a model after its first full training run: tuning a model that already exists on particular tasks, rather than building its general ability from scratch.
A Test Copy of Paiza's User Data Was Reachable From Outside for 13 Years, the Company Says
Japanese engineering recruitment company Paiza says up to 902,576 personal data records sat in a state where people outside the company could reach them for about 13 years. It reports no confirmed case of any of that data leaving its systems. The records were in a staging environment, a copy of the live service kept for testing. In a notice dated Nov.
A Navigation Test That Asks Who Is Walking
Put a low temporary barrier across a sidewalk and you have created three different problems. A robot dog walks around it. A blind pedestrian needs to be told about it, in words that don’t rely on sight. Someone in a wheelchair may have to turn back. The barrier is the same object each time. Whether the route can be used, depends on who is traveling it.
Tor Project Patches Its Network Defenses and Tells Relay Operators to Update
The Tor Project released version 0.4.9.17 of its software on Nov. 13, 2026, with a set of security fixes it described as a rapid response to a denial-of-service attack that was still hitting the network. It asked anyone running a relay, one of the volunteer servers that carry Tor traffic, to upgrade as soon as possible.
The Justice Department announced on Nov. 12, 2026, that it has seized cloud infrastructure used by North Korean hackers in a cryptocurrency theft scheme the FBI calls FaceStealer. In a separate action, it said it has court authorization to seize or freeze more than $1.2 billion in cryptocurrency that North Korea’s military stole from the Binance exchange.
AI Agents Asked to Predict Lab Results Agreed With Each Other More Than With the Lab
A research team has released a benchmark that hides the measured results of 4,207 real laboratory experiments and asks AI agents to predict them. In a preprint posted Nov. 11, 2026 and not yet peer reviewed, the team reports that thirteen configurations of frontier models agreed more closely with one another than with the measurements. The benchmark, called the Science Prediction Universe, is aimed at the question a scientist faces before spending bench time.
Check Point Firewall Flaw Is Already Being Used in Attacks, US Cyber Agency Says
The U.S. Cybersecurity and Infrastructure Security Agency added a flaw in Check Point security products to its Known Exploited Vulnerabilities Catalog on Nov. 13, 2026, on evidence that it is already being used in attacks. The flaw, CVE-2026-123039, lets a user who already has an ordinary account on one of the appliances take complete control of the device.
Trezor Says Three Wallets Sold by One Reseller Hid a Board That Copies the Screen
Trezor said three of its Safe 5 hardware wallets, devices that hold the keys to an owner's cryptocurrency, were sold by the reseller CryptoBilis with a hidden circuit board that copies everything shown on their screens. The company added that a firmware update released Nov. 11, 2026, stops the implants it has found from capturing anything on devices that install it. Trezor's embedded team analyzed the implant's software from the device reported in October.
MIT Researchers Trained a Climate AI on One Scenario They Designed Themselves
An MIT-led team has built a way to design the training data for climate emulators, the machine-learning stand-ins that substitute for full-scale climate models. Writing in Proceedings of the National Academy of Sciences, the authors report that an emulator trained on one scenario that their method produced outperformed an emulator trained on six standard emissions pathways, and did so on a smaller training set.
Apache Patches a Critical Web Server Flaw That Goes Back to 2023
The Apache HTTP Server project released version 2.4.71 on Nov. 11, 2026, fixing a flaw in mod_http2, the module that handles HTTP/2 connections. The project rates the flaw critical and says it can lead to remote code execution. The flaw is listed as CVE-2026-119618 and affects Apache HTTP Server 2.4.55 through 2.4.70, according to the project's advisory, whose release list dates 2.4.55 to January 2023.
Google Releases Gemini 4 Carbon, With a Mode That Puts a Team of Agents on One Prompt
Google made its newest Gemini model, Gemini 4 Carbon, generally available on Nov. 10, 2026, to paying developers, enterprises and its top-tier consumer subscribers, and is previewing a mode in its Antigravity coding tool that turns a single prompt into a team of software agents. Google says it built Carbon for the work that pushes a model hardest: complex, multi-step coding and debugging, and open-ended scientific research.
US Cyber Agency Says Attackers Are Already Exploiting an OpenSSH Flaw
The U.S. Cybersecurity and Infrastructure Security Agency added four vulnerabilities to its Known Exploited Vulnerabilities catalog on Nov. 9, 2026, citing evidence that all four are being exploited. One of them is in OpenSSH, the software widely used to log in to Linux servers remotely, though it affects only certain configurations.
A Tiebreak Rule for the Rare Cases Prison Risk Models Cannot Call
Two scoring systems can be built from the same prison records, tested on the same releases, and come out equally accurate. Then they can look at the same person and return different numbers. Whichever one is installed, its number goes into a file that prison psychologists, social workers and lawyers read, and that a court may later weigh when it rules on parole.
Three Men Charged With Mimicking Executives' Voices to Steal $129 Million
Federal prosecutors in Manhattan have charged three men with mimicking the voices and images of company executives to take more than $129 million from companies around the world. The U.S. Attorney's Office for the Southern District of New York announced the unsealing of the indictment on Nov. 5, 2026. It named Jacob Soall, Sharon Hazan and Guy Sheetrit, all arrested in Spain and the United Kingdom in 2026, and said extradition requests for them are pending.
A Language Model Edits Designed Molecules Until Software Can Plan a Route
Ask a generative model for a molecule that will fit a particular protein and it will hand back thousands, each one written out as a tidy string of letters and brackets. Give that list to a medicinal chemist and much of it comes back untouched. The shapes are plausible, the predicted binding is attractive, and nobody can see where to start building them. The field's own phrase for this is synthetically inaccessible: real on a screen, and stranded there. A study published on Sept.
Keeping AI From Forgetting Us Needs More Human Text Than the Math Implied
Feed a model its own output, generation after generation, and the answers start to thin out. The rare words go first, then the odd turns of phrase, until what comes back is a smooth average of itself. The failure has a name, model collapse, and the obvious remedy has been known almost as long: keep stirring fresh human writing into the training mix. What nobody has been able to say is how much.
Show More
loading
Actions
Is this you?
As a journalist, you can create a free Muck Rack account to customize your profile, list your contact preferences, and upload a portfolio of your best work.Get in touch with Wilkens
Contact Wilkens, search articles and posts on X, monitor coverage, and track replies from one place.
Learn more about Muck Rack