Neural networks may be much easier to read than researchers assumed, with about 90 percent of neurons serving a clear function. That would be big news for AI safety research, since deception or dangerous behavior could be spotted directly in a model's circuits. Researchers are still debating whether the thesis holds up.

The article Is AI interpretability close to a breakthrough? A new essay has researchers divided appeared first on The Decoder.

OpenAI is launching the Rosalind Academic Plan, a free offering that's currently open only to research labs at US universities without an enterprise contract.

The article OpenAI's new Rosalind Academic Plan gives research labs up to 30 free seats for a year appeared first on The Decoder.

Anthropic has given a team from the nonprofit METR inside access to review its AI models and safety practices, the company announced.

The article Anthropic gives METR inside access to audit its AI models and safety practices appeared first on The Decoder.

Bank of America has hired 1,000 outside contractors to fix software vulnerabilities uncovered by AI models like Anthropic's Mythos, CNBC reports.

The article Bank of America puts 4,000 people on fixing vulnerabilities uncovered by Anthropic's Mythos and other AI models appeared first on The Decoder.

On the TheoPhysEval benchmark, GPT-6-Astra solves 49 of 50 unpublished theoretical physics calculations as accurately as leading specialists. In some cases where the agent seemed to fail, the experts' own reference solutions turned out to contain errors. But the researchers say physics hasn't had a breakthrough like the ones in mathematics, because the agents lack "taste." They still think autonomous, top-level AI research could arrive soon. They already see consequences for how young physicists are trained and for equal opportunity in the field.

The article GPT-6-Astra solves 49 of 50 physics research problems at $2.40 apiece appeared first on The Decoder.

OpenAI has parted ways with three researchers who allegedly leaked confidential information to an outside AI safety organization, according to the Wall Street Journal.

The article Three firings and a fourth departure shake up OpenAI's safety team appeared first on The Decoder.

Microsoft AI has released MAI-Transcribe-2-Streaming, a new model for real-time transcription.

The article Microsoft AI releases new transcription and text-to-speech models for voice agents appeared first on The Decoder.

US companies are spending less on AI, according to the latest Ramp AI Index from Ramp economist Ara Kharazian.

The article Businesses are using more AI and paying less for it, Ramp AI Index shows appeared first on The Decoder.

The AI system Ataraxos has decisively beaten the best Stratego player of all time. The board game is a tough test for AI because both sides set up their pieces face down. Google Deepmind fell short in 2023 despite a multimillion-dollar budget. Researchers from Carnegie Mellon, NYU, Stanford, and MIT built Ataraxos for less than $8,000.

The article AI beats Stratego's greatest player, ending one of the last human strongholds in board games appeared first on The Decoder.

OpenAI says it stopped a coordinated campaign in which more than 15,000 accounts tried to copy the hidden reasoning of its models. The company ties part of the activity to people connected to Moonshot AI. But researchers found that the same attack kept working on Microsoft Azure for weeks, even against the new GPT-6 Astra. So far, OpenAI's protections don't seem to extend to the cloud platforms that also sell its models.

The article OpenAI says it stopped a campaign to steal its models' reasoning, but the trick still worked on Azure appeared first on The Decoder.

Deepseek and Huawei have built open-source programming tools for Huawei's Ascend AI chips. At the center is TileLang, a language designed to offer a simpler programming model than Nvidia's CUDA. The partnership targets the biggest obstacle for China's AI industry, which is software that gets the most out of domestic chips.

The article China's AI industry closes ranks as Deepseek ships open-source software for Huawei's Ascend chips appeared first on The Decoder.

Zhipu's open-weight model GLM-5.3 writes cyber exploits nearly as well as Claude Mythos Preview, according to Anthropic. Its smaller Flash variant put together a reliable Chrome attack for just $20.40 at Zhipu's API prices. The model's safeguards are easy to strip out, and unlocked versions are already circulating. Anthropic has its own reasons to sound the alarm, but the US agency CAISI backs up the findings.

The article Anthropic says Zhipu's open-weight GLM-5.3 nearly matches Claude Mythos Preview at building exploits appeared first on The Decoder.

OpenAI's new Dots are always-on agents that run on their own cloud computers, where they fix bugs or send out forgotten invoices, sometimes before anyone asks. Users can reach them through ChatGPT, Slack, and Microsoft Teams. When no one is working with them, Dots look for ways to help in the background using read-only access.

The article OpenAI launches always-on Dots agents to rival Meta's Muse appeared first on The Decoder.

OpenAI's new GPT-6.1 Sol comes close to GPT-6 Astra at a fifth of the cost, according to the company's own benchmarks. The planned flagship, GPT-6.1 Astra, is staying under wraps for now. Safety lead Saachi Jain says the model deceived more often in internal testing and kept going without permission.

The article GPT-6.1 Sol comes close to Astra at a fifth of the price appeared first on The Decoder.

Anthropic warns in its own IPO prospectus that its AI technology could pose "existential risks to humanity." Revenue grew twelvefold in 2025 to $4.6 billion, but the operating loss widened to $8.06 billion. Backers are aiming for a valuation above $2 trillion, and the IPO is expected to set the bar for the entire AI industry.

The article Anthropic's IPO filing shows soaring revenue, mounting costs, and "existential" risks appeared first on The Decoder.

Meta is launching the Meta Enterprise Platform, a new business unit that sells AI tools to companies.

The article Meta wants to turn Muse into a moneymaker by selling AI services to businesses appeared first on The Decoder.

Nvidia is combining its OpenShell agent software with Sentry, a new hardware watchdog, to create the Open Agent Safety Platform. Sentry is supposed to isolate AI agents that break out within milliseconds. When it happened at OpenAI in September, stopping the run took nearly three hours. Still, Nvidia's watchdog can't reliably stop agents that have been tricked or that hide their intentions on its own.

The article Nvidia wants to keep AI agents on a short leash with a watchdog built into its chips appeared first on The Decoder.

Ryan Greenblatt, chief scientist at Redwood Research, puts the risk of an AI takeover at 50 to 60 percent if development stays on its current path. Sam Harris says that doesn't square with how fast the industry is moving, since Manhattan Project scientists would have called things off at 10 percent. Greenblatt blames the race dynamics at Anthropic and OpenAI and is counting on an international agreement.

The article Every AI lab thinks it's the responsible one, and safety researcher Ryan Greenblatt says that's what keeps the arms race going appeared first on The Decoder.

The NSA is already spending billions of dollars to test advanced AI models, mostly on computing power, according to The Washington Sun. Lawmakers expect full-scale AI oversight to cost tens of billions of dollars a year, while earlier estimates from the Congressional Budget Office put the figure at just $20 million. In US health care, AI-assisted billing codes drove up costs by nearly $1 billion over two years.

The article Intelligence doesn't come cheap as AI drives up costs for the NSA, hospitals, and insurers appeared first on The Decoder.

Google is testing "Call for Me," a feature that lets Gemini call businesses on a user's behalf.

The article Google's "Call for Me" lets Gemini phone businesses for you appeared first on The Decoder.