Newsletter

FBI investigates reported breach of the criminal database police rely on

US media say the bureau is examining an intrusion into the network that holds warrants and background checks, as Geneva talks on autonomous weapons end with two more years of study.

A long concrete office building with projecting upper floors and rows of dark windows, seen from across a wide city intersection lined with trees.
Lead image The J. Edgar Hoover Building in Washington, D.C., headquarters of the FBI. The bureau has not commented on the reported intrusion. · Photo: ajay_suresh / Wikimedia Commons · CC BY 2.0
On this page

The FBI is investigating an apparent breach of the National Crime Information Center. The database is how federal and local police share warrants and background-check results. US media reported the inquiry on Saturday, citing two people briefed on the matter. The bureau did not respond to a request for comment from the BBC.

President Donald Trump was asked about the breach in a Fox News interview on Saturday. He said he had just heard of it. “I will say that it’s not good. I don’t like that,” he said. Neither the bureau nor the outlets citing officials have described what was taken or how an intruder entered.

The report lands in a difficult month for federal systems. Banking and air traffic control had outages last week that officials said were not cyberattacks. NBC reported a separate attempt on the Agriculture Department’s farm payment system. In September, the FBI said it was investigating a claim by ShinyHunters that it had stolen data on all 38,000 bureau staff.

Featured source: BBC , The Verge .

Geneva talks on autonomous weapons end with two more years of study

Diplomats at the United Nations in Geneva finished a week of negotiations at 4:30 a.m. on Saturday with a compromise. Countries would get two years to propose rules for lethal autonomous weapons, which could be binding or optional. The International Committee of the Red Cross had asked for binding rules within a year, with backing from the UN secretary-general. Russia objected to the agenda on the first day. The United States proposed a five-year window for non-binding guidelines.

The concern behind the talks is no longer theoretical. The UN says drones caused 80 percent of civilian deaths in Sudan this year. People still make the decision to kill in the systems now deployed. Humanitarian lawyers warn that software will soon choose targets alone.

Filed from: NPR , The Verge .

Model projects that GLP-1 medicines could free farmland the size of Italy

Researchers at Aberdeen and Edinburgh modelled what widespread use of GLP-1 obesity medicines would do to food demand and land through 2050. They assume use spreads among adults living with obesity while diets hold steady. On that basis they estimate up to 6.6 million deaths avoided each year and 32 million hectares more natural land. Adding a shift toward the EAT–Lancet reference diet raised the figures to 7.7 million deaths and 46 million hectares.

The paper also describes a rebound. Cheaper food could encourage more eating among people who take no medication, including of unhealthy products, which would erode part of the gain. These are simulation results. They rest on assumptions about uptake and about how farmers and food companies respond.

Filed from: Nature Communications , The Conversation .

A microdrama startup pays an actor to license his face for an AI series

Shortical premiered a fully AI-generated series in Los Angeles last week. “Inevitable” is built around a digital avatar of the actor Aki Avni. He plays a version of himself in a thriller about an actor who sells his likeness to a studio. It is the first time the company, founded in January 2025, has built a series around a real person’s face. Shortical also opened a portal. People submit social media profiles for review and, if accepted, get a contract naming the series, the character and the payment.

A company representative said the portal has produced no signed agreements, and that Avni’s deal was arranged directly. Neither disclosed the sum. The representative said Avni came on as a partner and was paid in line with a comparable non-AI acting role. Rest of World reported this year that Chinese microdrama firms pay between $15 and $700 for facial likenesses.

Filed from: TechCrunch , Shortical .

From the Community

Lean community certifies a machine-checked proof of the Riemann Hypothesis

The Lean and Mathlib communities published a joint statement giving full confidence to a proof of the Riemann Hypothesis. Levent Alpöge released it on November 20 as a 17-million-line Lean library. The communities report that the code compiles. They say two comparison tools confirmed the theorem matches Mathlib’s own statement, and that the exported proof passed Lean’s kernel and independent checkers. Two of the checkers carry formal proofs, relative to set theory, that they cannot accept a false proof.

The verification took about a day. The statement says the mathematical ideas in the proof still need to be explained. It is a community certification, not peer review by number theorists.

Filed from: Lean community .

Analysis estimates Anthropic’s US data centers will run mostly on gas and coal

The Better Data Center Project, which studies the power demand of AI infrastructure, published a case study of Anthropic’s compute contracts. It estimates that about 87 percent of the company’s US data center power will come from gas in 2028, and 11 percent from coal. It also estimates that Anthropic could emit 32 million metric tons of carbon dioxide that year. The group says the company has disclosed renewable commitments in Norway and Canada but none in the United States.

The figures are the project’s own estimates. They are built from announced sites and assumptions about which power plants meet new load. Anthropic has not published a comparable breakdown.

Filed from: Better Data Center Project .

Ghostscript font bug let uploaded files run commands in preview pipelines

Researchers at pwn.ai published details of CVE-2026-125396, a use-after-free in Ghostscript’s Type 1 font loader. A crafted document can escape the renderer’s sandbox and run commands as the process that opened it. The firm reports confirmed execution through Google’s document pipeline, WordPress uploads, a Debian 13 print queue, Roundcube attachment thumbnails and Plesk. Ghostscript 10.08.0 carries the upstream fix, and Debian shipped an update.

WordPress issued a separate fix, CVE-2026-65640, for its upload path. The affected-product list and timeline come from the researchers, and several of their case studies remain unpublished.

Filed from: pwn.ai .

Continue reading

Complete index →