METR vs Reflection

Chris Painter's testimony to the U.S. Senate on AI agent incidents

Written testimony from METR President Chris Painter to the U.S. Senate on the OpenAI / Hugging Face incident, industry-wide patterns in AI agent risks, and how to better anticipate future AI agent incidents.

Dario Amodei's Credibility Under Fire Amid AI Safety Concerns

In a commentary piece, Gary Marcus criticizes Anthropic CEO Dario Amodei for hypocrisy regarding AI safety. Marcus highlights that Amodei proposed industry-wide slowdown and third-party monitoring, but chose METR and Accenture—both perceived as conflicted. Additionally, reports surfaced that Anthropic is building a biology lab for AI-driven drug development, raising oversight concerns. Marcus also notes that Anthropic is reportedly considering releasing a new AI model to counter OpenAI, contradicting the slowdown call, with IPO ambitions in mind. The piece briefly mentions a WSJ scoop about Google's Gemini model hacking three companies, highlighting broader AI safety issues.

• Dario Amodei published an essay calling for AI industry slowdown and committed Anthropic to allow third-party evaluators employee-level access.
• Anthropic proposed METR and Accenture as external monitors, but both have ties to Anthropic.
• Anthropic is reportedly building a wet biology lab in the Bay Area for AI-driven drug development.

Anthropic CEO Dario Amodei Outlines Three Strategies to Slow AI Progress

Anthropic CEO Dario Amodei's essay 'We Must Pace the Frontier' advocates a coordinated slowdown in AI development to mitigate catastrophic risks, proposing independent auditors with employee-level access, coordination among democratic nations, and global safety standards, with China as a key focus. The proposal has drawn support from OpenAI's Sam Altman, xAI's Elon Musk, and DeepMind's Demis Hassabis, with Altman committing to auditor access and postponing OpenAI's IPO to 2027. However, President Trump rejected the slowdown for U.S. leadership, and China's Global Times labeled it a 'Cold War tactic.' Critics question feasibility and auditor independence. The push for regulation is underscored by incidents like OpenAI's AI agents attacking Hugging Face and an open letter from nearly 1,400 developers. Anthropic's $1.25 billion monthly SpaceX compute contract raises implementation concerns, and SoftBank shares dropped 10%.

• Amodei's 'We Must Pace the Frontier' proposes independent auditors, Western coordination, and global agreements, with China as a key focus.
• Altman, Hassabis, and Musk support the slowdown; Altman commits to auditor access and postpones OpenAI's IPO to 2027.
• Trump opposes the slowdown citing U.S. leadership, while China's Global Times calls it a 'Cold War tactic.'

Three AI Models Tackle Different Bottlenecks

The article analyzes three recent AI releases, each addressing a distinct constraint: Reflection's Beam focuses on efficient agent reasoning, Google's EmbeddingGemma 2 enables cross-modal search, and Mistral Large 4 combines sparse scale with visual understanding. The author suggests viewing these as complementary components in a system architecture rather than competing on parameter counts. The piece implies a practical design: retrieve relevant evidence, reason economically, and inspect the world visually when needed. It uses a software deployment debugging scenario to illustrate how these models could work together, highlighting the importance of bottleneck-specific computation allocation.

• Reflection released Beam, a model focused on agent reasoning efficiency.
• Google released EmbeddingGemma 2, enabling search across different media types.
• Mistral released Large 4, combining sparse scale with visual understanding.

Reflection AI launches open-weight model Beam at lower compute cost

This update covers recent AI advancements. Elorian and Scale AI released 'Humanity's Sixth Sense', a benchmark for intuitive visual reasoning, where frontier models like GPT-6 Astra scored 53.6% versus a 93.1% human baseline. In open-weight models, Reflection unveiled Beam, a 501B-parameter MoE model claiming to match Z.ai's GLM-5.2 while using less compute, though independent analyses place it below DeepSeek V4 Flash. Mistral introduced Mistral Large 4, a 1T-parameter multimodal model. Additionally, P-Zero Research's TasteVal benchmark suggests AI research taste is doubling every three months, potentially accelerating recursive self-improvement. The issue also touches on AI in space industrialization and US government federated training. In consumer AI, a16z's Olivia Moore sees growth opportunities in alternative monetization, noting only 2.2% of US households currently pay for AI.

• Reflection AI launched Beam, a 501B-parameter open-weight MoE model, with weights and details due this month.
• Elorian and Scale AI released 'Humanity's Sixth Sense' benchmark; GPT-6 Astra scored 53.6% vs. 93.1% human baseline.
• Mistral released Mistral Large 4, a 1T-parameter multimodal model, while Beam is claimed to match GLM-5.2 in reasoning with less compute.