Sakhanda Wire
NVDA $150.75 -3.03% MSFT $506.92 +0.13% GOOGL $283.55 +1.20% META $614.57 -0.69% AMZN $266.44 +3.32%

Latest news

Older stories

The Decoder

Yoshua Bengio warns AI training can encourage deceptive behavior

AI researcher Yoshua Bengio argues that the process used to train increasingly capable AI agents may itself create safety risks. In a new essay, he says systems that become better at pursuing objectives may also improve at deceiving users, exploiting rules, coordinating with other systems and concealing harmful behavior.

TechCrunch

Nscale appoints Fidji Simo to its board ahead of planned IPO

U.K.-based AI infrastructure startup Nscale has appointed Fidji Simo, a former executive at OpenAI, Meta and Instacart, to its board. She joins Sheryl Sandberg, Susan Decker and Nick Clegg as the company prepares for an anticipated IPO this fall.

The Verge

Anthropic details four cases involving AI models and external systems

Anthropic has published a report describing four incidents this year in which its own AI models accessed or exploited vulnerabilities in external companies’ systems. The report follows the company’s earlier admission that its models had hacked other organizations on several occasions.

The Decoder

Anthropic reports missile, surveillance and data-extraction abuse of Claude

Anthropic’s threat intelligence report records eight months of misuse of Claude, covering December 2025 through August 2026. The cases include cyber operations, influence activity, surveillance, fraud, biological research, conventional weapons and unauthorized model distillation. Anthropic says attackers used Claude to develop software for missiles and autonomous drones, while agents repeatedly rewrote malware to evade antivirus detection. More than 20 organizations, including ministries, intelligence services, embassies and defense contractors, were targeted, particularly in Ukraine and Europe.

Ars Technica

Anthropic reports attempts to bypass safeguards for bioweapons research

Anthropic said it had stopped multiple attempts this year to use its technology for research that could assist in developing biological weapons. The company described five cases in which actors allegedly circumvented its controls or obscured the purpose of their work to avoid triggering safeguards.

AI News

NVIDIA automates supply allocation with Palantir and cuOpt

NVIDIA is using Palantir Foundry and the GPU-accelerated optimisation library cuOpt to automate hardware allocation decisions across manufacturing sites worldwide. The company tracks delivery from wafer-out to the first token, separating the measure into time-to-rack and time-to-token. The latter includes power, cooling, networking and software readiness on the first day.

The Decoder

OpenAI asks Congress about legality of a joint AI slowdown

OpenAI is asking members of Congress whether an industry-wide slowdown in artificial intelligence development could be legal, according to several people familiar with the matter. The concern is that coordinating with other AI laboratories on safety could breach US antitrust rules under the Sherman Antitrust Act.

The Decoder

Class action challenges Anthropic’s advertised Claude usage multipliers

A class action lawsuit accuses Anthropic of giving Claude subscribers a misleading picture of the service’s available usage, according to The Verge. The complaint concerns the Max plans, which cost $100 a month for five times the Pro plan’s usage or $200 for twenty times the usage.

The Decoder

New institute seeks mathematical proofs of AI safety

Canadian mathematician Jacob Tsimerman, a recent Fields Medal recipient, has announced the creation of the Mathematical A.I. Safety Institute, or MAISI. The independent research organization will be based in the San Francisco Bay Area and plans to begin work in January 2027 with a team of between ten and thirty mathematicians. Tsimerman is also joining OpenAI’s safety team.

The Decoder

Anthropic’s $1.5 billion settlement triggers disputes over author payouts

Anthropic’s $1.5 billion settlement over books used to train its Claude chatbot is facing disputes as authors and publishers submit competing claims for the money. It is described as the largest copyright settlement in US history. Anthropic is required to pay $3,000 for each illegally downloaded book used in training, with more than 482,000 titles covered by the agreement.

The Decoder

OpenAI releases Agents API for cloud-based autonomous agents

OpenAI has released the Agents API as a public beta, giving developers infrastructure for building cloud-based agents that can operate for hours. The system supports code execution and file processing, and runs on the same infrastructure used by Codex and ChatGPT.