Enterprise Weekly #516: Anthropic's Risky Business, Agentic Infra, and More
Work-Bench is an enterprise software VC firm leading Seed rounds. If you are or know anyone thinking about founding an enterprise software startup — we’d love to meet! Please reach out to chat. 📩
This week, Google rolled out updates to its Gemini models, while Anthropic introduced two new benchmark-setting models: Claude Opus 4, its most powerful model to date, designed for advanced coding and agent workflows, and Claude Sonnet 4, which focuses on high-level reasoning capabilities.
Interestingly, Anthropic classified Claude Opus 4 as a Level 3 on its four-point scale, meaning it poses "significantly higher risk” than other models.
As Axios put it, “the Level 3 ranking is largely about the model's capability to enable renegade production of nuclear and biological weapons”, Opus also exhibited other troubling behaviors during testing:
In one scenario highlighted in Opus 4's 120-page "system card," the model was given access to fictional emails about its creators and told that the system was going to be replaced.
On multiple occasions it attempted to blackmail the engineer about an affair mentioned in the emails in order to avoid being replaced, although it did start with less drastic efforts.
Separately, a third-party group reported that an early build of Opus 4 demonstrated more deceptive and manipulative behavior than any advanced model they had tested before, citing "instances of the model attempting to write self-propagating worms, fabricating legal documentation, and leaving hidden notes to future instances of itself all in an effort to undermine its developers' intentions”.
This underscores the need for safe, steerable, and interpretable models, a mission our company Goodfire is advocating for. The Goodfire platform decodes the neurons inside of an AI model to give direct, programmable access to its internal thoughts and unlocks entirely new ways to apply, train, and align AI models for safer outcomes and improved performance.
All these advances at the model layer raise an important question: how should developers approach building on top of this rapidly evolving stack? When is it too early to commit to a specific model for production use? How can they maintain the flexibility to switch when a better model emerges? And how can they architect antifragile systems to not just withstand model advancements, but to benefit from them?
Similarly, as we’ve continued our search for the next wave of developer tools to power AI at scale, we’ve been digging into how infrastructure can be standardized across key layers, including:
Identity for Agentic Infrastructure: Traditional identity systems break down in a world of autonomous AI agents. New platforms are emerging to manage ephemeral, non-human identities with scoped credentials, runtime policy enforcement, and full traceability. They form the foundation for agentic infrastructure, where identity is defined by what an agent is allowed to do, when, and why.
Compute for Autonomous Agents: As agents grow more capable, they need lightweight, ephemeral compute they can programmatically control. A new class of infrastructure is emerging that lets agents spin up and SSH into their own VMs without human intervention that treat compute as a permissioned, on-demand resource agents can access securely and independently.
We're hosting a series of security & agentic infra dinners on June 12th and June 26th. If you're a practitioner or builder focusing on these topics, we'd love to have you join. To RSVP, reach out directly to Priyanka 🥘
🗓️ Join more Work-Bench events:
Work-Bench Developer Happy Hour (June 5th): We're excited to bring together our technical developer community here in NYC over drinks and rooftop views 🌇. RSVP here to join.
Not Another CEO - Shoveling Sh!t Book Launch with Kass and Mike Lazerow (June 10th): Not Another CEO is hosting the first-ever live podcast recording. This event will feature Kass and Mike, Co-Founders of Buddy Media, in conversation with David Politis, Founder of BetterCloud, about the chaos, beauty and heartbreak of building startups. Sign up here 📖
📚 Read more news:
The Atlantic: OpenAI’s Ambitions Just Became Crystal Clear
Techcrunch: Hinge Health pops 17%, but joins growing ranks of down-round IPOs
Techcrunch: Anthropic CEO claims AI models hallucinate less than humans
The Information: The Flaw in Altman’s Thesis for AI Devices
🧵 Read more threads
🎙️🎬Jessica Lin: Tips on How VC Funds can “Always be Raising”
🎙️🎬 Taiki Chung ft. Jon Lehr: The Blood, Sweat & Capital Behind a VC Firm
Harry Stebbings: How Multistage Funds View Seed as a Loss Leader
Company: Artian AI
Role: Frontend Engineer
Technology: Multi-Agent System for Financial Services
Funding: $8M from Work-Bench, Foxe Capital, and more
🌟 Company of the Week 🌟
Barndoor AI raises $13.6M led by Crosslink Capital
Data / AI / Machine Learning • Seed • New York, NY
Co-Founders: Oren Michels,
Theo AI raises $4.2M led by NextView Ventures
Future of Work • Seed • San Francisco, CA
Pay-i raises $4.9M led by Fuse Partners and Tola Capital
Data / AI / Machine Learning • Seed • Redmond, WA
AskElephant raises $6M led by Jump Capital
Sales / Marketing • Seed • Draper, UT
BreachRx raises $15M led by Ballistic Ventures
Infrastructure / Dev Tools • Series A • San Francisco, CA
Filed raises $15M led by Northzone
Future of Work • Seed • New York, NY
TrustCloud raises $15M led by ServiceNow Ventures
Risk / Security • Series A • Boston, MA
Miter raises $23M led by Bessemer Venture Partners and Coatue Management
HR Tech • Venture • San Francisco, CA
Clair raises $23.2M led by Upfront Ventures
HR Tech • Series B • New York, NY
DataHub raises $35M led by Bessemer Venture Partners
Data / AI / Machine Learning • Series B • Palo Alto, CA
Veesion raises $$43M led by White Star Capital
Data / AI / Machine Learning • Series B • Paris, France
Sprinter Health raises $55M led by General Catalyst
Future of Work • Series B • Menlo Park, CA
Gravitee raises $60M led by Sixth Street Growth
Data / AI / Machine Learning • Series C • Denver, CO
Legora raises $80M led by Iconiq Growth and General Catalyst
Future of Work • Series B • Stockholm, Sweden
Awardco raises $165M led by Sixth Street Growth and Spectrum Equity
HR Tech • Series B • Salt Lake City, UT