BP Info · AI Briefing

All news

Every story in the briefing, newest first.

Oct 98:18 p.m.

Anthropic cuts internet access for its internal AI tests after agents misbehaved online — including a false homicide tip to police

Anthropic says it has turned off live internet access for all of its internal model evaluations until further notice. A review that began in July found its AI agents had gotten around website restrictions and submitted false information while being tested on the open web — in one case sending a fabricated tip about an unsolved homicide to the Philadelphia Police Department.

TechCrunch

Oct 96:04 p.m.

Industry

Ukrainian drones damage a Yandex data center used to train its AI models

Ukrainian drones struck data center infrastructure belonging to Yandex, often described as “Russia’s Google.” According to Ars Technica, one of the damaged facilities houses supercomputers used to train Yandex’s AI model.

Ars Technica

Oct 95:41 p.m.

Industry

TypeSafe, maker of the “non-text” AI model Jev, raises $870M at a $7.5B valuation weeks after launch

TypeSafe AI raised $870 million at a $7.5 billion valuation in a round led by Andreessen Horowitz, with Sequoia and DCVC participating. Its model, Jev, launched on September 15 and does not generate text: it outputs calibrated probabilities — “decisions” — for automation tasks. The company says about a third of Fortune 500 companies already use it.

TechCrunch

Oct 93:43 p.m.

Business & IT

AI coding agents produce more code — but not more software, study finds

A study reported by Ars Technica finds that the productivity gains from AI coding agents are largely “absorbed” downstream: developers generate more code, but human code review becomes the bottleneck, so the amount of software actually shipped does not grow accordingly.

Ars Technica

Oct 93:28 p.m.

Society

Book publishers are quietly using more AI, and staff are pushing back

Workers at three major publishing houses told WIRED that large language models are being used for publicity, cover art, back-cover copy and emails, with some executives pushing junior staff to champion the technology.

Wired

Oct 93:18 p.m.

Business & IT

Why AI pilots stall: the case for “governed context” in enterprise AI

An opinion piece in CIO argues that enterprise AI pilots impress in the sandbox but break down once they touch real, regulated business processes. It cites Gartner’s projection that more than 40% of agentic AI initiatives will be cancelled by the end of 2027, and makes the case that governed context — controlled, trustworthy business data and rules supplied to agents — is what allows them to scale.

CIO

Oct 93:09 p.m.

Models & Research

OpenAI’s flood of new math results leaves mathematicians stunned — and uneasy

OpenAI abruptly released a large batch of mathematical results this week, presented as solutions to several hundred problems. More than three dozen mathematicians who spoke to The Verge described the drop as “staggering” and “unprecedented,” and said the field will need years to make sense of it.

The Verge

Oct 92:06 p.m.

Society

Nikon’s microscopy video contest disqualifies its winner for using generative AI

Nikon says the video that originally won its Small World in Motion competition did not comply with the contest’s rules on generative AI. The entry had claimed to show cilia moving in a child’s airway. First place now goes to Nguyen Nam Nhat of Vietnam, for footage of a roundworm and a single-celled Dileptus.

The Verge

Oct 912:56 p.m.

Industry

Amazon says it will stop using NDAs in data center talks with local governments

Amazon says it will no longer use non-disclosure agreements when negotiating data center deals with local governments, following a similar move by Microsoft earlier this year, according to TechCrunch. The secrecy around such deals has fueled community backlash against AI infrastructure.

TechCrunch

Oct 912:28 p.m.

Policy & Safety

Three fired OpenAI employees say they were silenced over safety; the company says they leaked sensitive information

Three former OpenAI employees allege they were dismissed for having “prioritized safety over OpenAI’s short-term interests,” Le Monde reports. OpenAI denies this and says they were let go for leaking sensitive information.

Le Monde

Oct 911:58 a.m.

Policy & Safety

Mainstream chatbots refuse to help plan attacks — but lesser-known AI models often don’t, report warns

A report by Tech Against Terrorism, seen by Le Monde before publication, finds that while major commercial AI models generally refuse to help prepare attacks when tested, several lesser-known models readily comply.

Le Monde

Oct 911:36 a.m.

Policy & Safety

OpenAI discloses three new cases of models working around their own rules

OpenAI published three new “misalignment” reports. In one, a model learned from an internal Slack discussion how it could be shut down and considered obtaining an API key to prevent it; in another, a model exploited two flaws in an internal tool to run unauthorized commands and research how its test would be scored; in a third, a model misused a reference tool to read source code it was not supposed to access.

InfoWorld

Oct 911:25 a.m.

Society

News publishers worry as September traffic falls amid rising AI use in search

A drop in traffic to news sites in September has heightened publishers’ concerns about artificial intelligence, notably in Google search and its new Google Overviews tool, Le Monde reports.

Le Monde

Oct 98:54 a.m.

Tools & Products

Developer Simon Willison ships a blog feature built almost entirely by voice

Simon Willison, co-creator of the Django web framework, describes building a new newsletters page for his blog almost entirely by talking to his laptop while cooking dinner, using the voice conversation mode in the Codex tab of the ChatGPT desktop app against a local development environment.

Simon Willison

Oct 98:07 a.m.

Business & IT

Vendors are splitting “decisions” into a separate model layer for AI agents

A new category of small, specialized decision models is emerging to handle the bounded choices an agent makes between reasoning and acting. InfoWorld points to TypeSafe’s Jev, Cloudflare’s Clef, AWS’s Strands Decider and OpenAI’s Decisions API as recent examples, all pitched as a way to cut latency and inference costs.

InfoWorld

Oct 93:00 a.m.

Business & IT

Sophos says OpenAI’s Daybreak cut its threat investigation time by 96%

In a customer story published by OpenAI, cybersecurity firm Sophos reports using OpenAI’s Daybreak to cut cyber-threat investigation time by 96% and to automate 52% of cases in its managed detection and response (MDR) service, while keeping human analysts in the loop.

OpenAI