GOVERNANCE

Anthropic’s audit push tests AI transparency

As calls for safety standards and independent audits of AI models continue to mount, Anthropic has revealed its plan to open the door to a third party. 

On Thursday, the AI lab announced that it will give "employee-like access" to a six-person team from METR, a nonprofit that evaluates frontier model risks, allowing them access to company hardware, office space, and internal tools. The company said that METR will publish the results of its risk assessments, and it has worked with the nonprofit on previous reviews of its own risk reports. 

Anthropic noted that it's employees can share any information they would like with this team, other than a few specific categories, which include data about customers, attorney-client privileged information, or information about it's model architectures.

"Embedded evaluation gives a few qualified people working under Anthropic’s security standards a close view of how we operate," the company said in its blog post. "It’s not a way for other organizations to obtain our information, or that of our customers and partners." 

The announcement follows Anthropic committing to third-party evaluations of its models in CEO Dario Amodei's latest essay, titled We Must Pace the Frontier, as well as signing the White House Accord on Super Intelligence, a good-faith agreement which recommended a multi-layered approach to external audits. 

Anthropic also laid out its approach to working with these auditors, noting that it's giving them "deep access" to information to "share evidence and reasoned conclusions" about model risks. The company said that these evaluators require four things: 

• Frontier model capability expertise, such as the ability to identify when an evaluation goes wrong;

• Financial and operational independence from the company that its evaluating;

• "Enough access" to evaluate risk, including both the model itself and how the models were trained;

• And the freedom to publish their observations and assessments.

"We want to be publicly accountable for the decisions we’re making," Anthropic said in its blog post. "But we can’t publicly share all the information relevant to our safety work. Some of it could be useful to malicious actors, or is commercially sensitive, like compute or training information." 

Though rival OpenAI has not yet announced that it is giving a similar level of access to embedded evaluators, the company does already work with METR and other independent evaluators. 

Anthropic's announcement, however, also follows a Tuesday report from Reuters claiming that Anthropic and OpenAI are attempting to gain control over what evaluators can publish as a means of preventing IP exposure, potentially limiting the transparency of these evaluations.

Third-party evaluations are a crucial way to hold these labs accountable for the risks that their models present. But the effectiveness of these evaluators may only go so far. For one, if they are restricted in what they can publish, as the Reuters report suggests, then they aren't able to be fully transparent to the public. Additionally, because the existing standards and "morally binding" commitments are all practically based on the honor system, whether or not these labs will take the findings of the evaluators to heart is entirely up to the labs themselves. Though voluntary measures such as this are a positive step in the right direction for AI safety, real safety measures may require firmer, mandatory regulation that requires these companies to make changes when a risk is presented.

TOGETHER WITH MEMOKET

Never lose a meeting detail again

Back-to-back meetings. Coffee chat follow-ups. Half the details already gone?

Memoket changes that. Memoket Gem is the lightest AI wristband on the market — sitting alongside your Apple Watch, it records your conversations with one press and turns them into instant summaries, action items, and full context across every meeting. 

With Memoket Gem, you get:

• Every conversation captured hands-free, worn as a wristband, clipped to your Apple Watch, or as a pendant

• Context connected across conversations, so you can ask AI what happened, what changed, and what's next

• Seamless sync with Slack, Notion, ChatGPT, Claude, Gmail, and your calendar

• Write your work report, or project timeline, all based on your actual meetings, without prompting AI all over again. 

Get 10% off with code DEEPMEMO26 through Oct 31.

Get Memoket Gem →

BIG TECH

Meta, Nvidia give US open models a boost

In its second major model release this week, Meta is going back to its open source AI roots. 

On Thursday, the tech giant unveiled Muse Spark Open, which it's calling "the most capable US open-weight model to date," and fulfilling its promise it made in April to release the weights of Muse Spark. It represents the first meaningful contribution to the US open source market since the company's Llama models. 

Notably, Meta said that this is the first model of its size class capable of running on two Apple M5 Ultra Mac Studios with the latest MLX framework. The model was specifically optimized to run on Mac Studio with M5 Ultra, with Meta CEO Mark Zuckerberg specifically calling out the work of Apple CEO John Ternus in a tweet announcing the models. 

"We believe the future is for everyone, and that includes the freedom to run advanced models wherever, including at home," Meta said in the announcement. 

The model is natively multimodal, clocking in at 816 billion total parameters with 42 billion active parameters and a one million-token context window. Meta said it was built to handle coding and agentic work, as well as long-horizon tasks, complex software engineering and multi-step reasoning.

• In benchmarks for model context protocol, academic use, deep reasoning and hallucination, the model beats out competitors from Thinking Machines, Nvidia's Nemotron, Mistral, and Chinese labs like Moonshot, Z.ai and Alibaba. 

• However, Moonshot's controversial Kimi K3 still has it beat in the Terminal-Bench 2.1 and DeepSWE 1.1 benchmarks for coding and software development. 

Muse Spark Open is available today through several API platforms, including Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI, and model weights can be downloaded from Amazon, Google Cloud, Hugging Face and Microsoft. 

But Meta isn't the only company seeking to claim its piece of the US open source ecosystem. Also on Thursday, Nvidia unveiled Nemotron 3.5 Super VL, an open, multimodal model built specifically for complex, agentic enterprise workflows. The model features 120-billion parameters with 12 billion active, which balances "model capacity with inference efficiency," Nvidia said in its announcement. 

The model is customizable for domain specific data, tools and policies, and performs well across coding, web research, document intelligence, cybersecurity and video understanding tasks, the company said. The model weights are available today, as well as the open datasets on Hugging Face.

In the US, the open source AI ecosystem is still finding its feet. But increasingly, enterprises and organizations are turning to open source models as a cost effective alternative to the expensive proprietary APIs that companies like OpenAI and Anthropic are pushing. Though Nvidia and Meta likely have different motivations for wanting to feed the open source ecosystem, with Nvidia's strategy largely driven by designing its hardware, the result is still more options for enterprises who want to save money without wanting to take on the security risks that Chinese open source models may present. 

TOGETHER WITH CODER

The Secret To Safe AI

AI isn’t inherently dangerous – but as we’ve seen across the entire industry with multiple labs’ recent testing leaks, even the most secure environments can still be vulnerable.

So, how’s a company supposed to scale their AI adoption while also protecting themselves? One word: Control. And that starts with updating your existing infrastructure to include a layer dedicated specifically to AI. Coder calls this the “AI Operating Layer”, and this free whitepaper unpacks what it does, why it matters for your business, and the 5 things it enforces that will help keep your company secure.

Check out the free interactive whitepaper right here.

CONSUMER

Amazon’s edge in AI wearables is its ecosystem

From smart wearables to AI-enhanced apps, using AI to log your days has become one of tech's hottest trends. Now Amazon is joining the race. 

On Thursday, Amazon announced the Bee Frontier and Bee Eclipse, two personal AI gadgets that can listen to you throughout your days to provide summaries, to-do lists, reminders, and scannable directories of your day. While these devices are essentially a take on the AI pin offered by rival Plaud, Amazon is relying on its integration with Alexa as well as a lower price point to give it a competitive edge. 

The Bee Frontier can be worn as a pin or as a wristband and was designed to be worn constantly, featuring seven days of battery life, IP67 water resistance, and two microphones with noise filtering that allows you to capture audio anywhere you go. This product is the next iteration of the Bee Pioneer, the startup's debut product that was also an AI note-taking wristband, before it was acquired by Amazon in July of 2022. 

The Bee Eclipse serves the same purpose as the Bee Frontier with a different form factor: a slim, lightweight device that attaches to the back of your phone magnetically. It has more than 21 days of battery life, and  a visible indicator when it is summarizing a conversation. While other competitors, such as Plaud, have offered the same products before, there are some key differentiators: 

• Price: The Bee Frontier is $74.99 and the Bee Eclipse (available soon) is $99.99. Meanwhile, the Plaud Pin is $179 and the Plaud Note is $159. 

• Alexa integration: With every purchase, users receive conversation summaries, recaps, and 14-day memory at no additional cost. However, with a Bee Pro subscription of $11.99, users get perks such as integrations with Amazon.com and Alexa+, which allow users to ask Alexa things such as "What are my Bee insights from yesterday?” or “What are my Bee to-dos?”

Of course, an always-on recorder like this raises obvious privacy concerns. Much like Apple’s approach to always-on listening with the S11 chip in the Apple Watch, Bee Private Compute uses a security architecture “that processes your conversations inside a secure and isolated processing environment where no one, including Amazon, can access your data,” according to the release. On top of that, audio isn’t stored, and users can turn off recording whenever they’d like and delete their transcripts.

When I first tried the Bee Pioneer in January 2025, what surprised me most was how genuinely useful its insights were. Being able to look back and reflect on the day’s interactions felt valuable on its own, and getting feedback on my actions was an added bonus. Still, two barriers have always held me back from devices like this: the upfront cost of hardware such as Plaud’s NotePin, and the siloed software they run on. If I want these kinds of in-depth insights, I don’t want to open yet another app or deal with a device that can’t connect seamlessly to the ecosystem I already use. That’s why the Alexa+ integration feels so useful. Many people already own an Echo speaker or smart display, and now they can simply ask it for these insights. Even something as simple as “When did my partner say he was coming home on Tuesday?” is helpful.

LINKS

• Anthropic reportedly made $16.8 billion in Q3 revenue

• Eight states sue OpenAI over alleged failure to protect vulnerable users

• OpenAI agrees not to use trade secrets claimed stolen from Apple

• North.cloud raises $60 million in Series B funding

• Cohere to acquire AISLE and launch Cohere Cyber unit

• Bank of America hires 1,000 contractors after Mythos found flaws

• Grok Voice 3.0: "the world’s most intelligent speech-to-speech model"

• OmniHuman 2.0: BytePlus model to bring audio and images to life

• Claude Code desktop app: Updated to make it easier to start session, split work, etc

• Perplexity Computer: Max subscribers get Ultrafast versions of GPT-6 Astra, SOL 6.1

• Code Review: out of research preview, available to Claude Team and Enterprise

• Genesis AI: Data Annotation Specialist, Physical AI

• TikTok: AI & Data Architect - Trust and Safety

• Contivio: AI Communications Solutions Engineer

• Funko: AI Solutions and Enablement Analyst

GAMES

A QUICK POLL BEFORE YOU GO