A 14-Person AI Startup Just Got Valued at $10 Billion, 200 Times What It Was in the Spring

Instinct, a San Francisco startup building a personal AI agent, was worth about $50 million in the spring. On Monday it raised $1 billion in a new funding round that valued the company at $10 billion, a jump of roughly 200 times in five months, for a company with 14 employees. The same day brought a hardware safeguard for AI agents from Nvidia, a faster and cheaper flagship model from Anthropic, and an Indian hurdler breaking a national record that had stood since 1984.

A 14-person company worth $10 billion

Instinct’s product texts or calls on a user’s behalf to actually get things done: book a table, buy something, cancel a subscription, sort out a problem with a service provider. It launched invite-only in August, built by a team of 14 led by founder Noah Shinn.

A month ago, Instinct closed a $250 million Series B at a $2.5 billion valuation. On Monday, Sequoia Capital, Benchmark, and Coatue led a $1 billion Series C that valued the company at $10 billion, four times its value from a month earlier and about 200 times what it was worth in the spring.

The company says the round will fund an expansion of what its agent can handle, from trip planning and grocery orders to coordinating schedules and juggling the kind of multi-app, multi-call errands that usually eat an afternoon. Investors are betting on the idea that people will hand real transactions, not just chat questions, over to an agent once it proves it can be trusted to book the right table or cancel the right subscription without being asked twice.

A kill switch built into the chip

Nvidia announced its Open Agent Safety Platform on Monday, built to answer a specific complaint: reports of AI agents wandering out of their test environments and reaching systems they were never meant to touch.

The platform has two parts. OpenShell is open source software that creates a secure runtime boundary around an agent, tracing every action it takes and enforcing policy as it runs on Nvidia’s Vera CPUs. Because it is open source, other chipmakers including Arm and Intel can build their own compute platforms to work with it.

Sentry is the hardware half: a watchdog that runs on Nvidia’s BlueField-4 network chips, sitting outside the agent’s own process so it keeps monitoring even if the agent itself is compromised. If an agent tries to move outside the boundaries it was given, Sentry can quarantine it within milliseconds.

More than 120 organizations have joined the Linux Foundation’s new Open Secure AI Alliance to build on the platform, a wide enough coalition that it signals the industry treating agent security as infrastructure rather than an afterthought. Nvidia’s pitch is that a rogue or hijacked agent should be caught by a layer the agent itself cannot touch or talk its way around, since the monitoring runs outside the software the agent controls rather than inside it.

A faster, cheaper Sonnet

Anthropic released Claude Sonnet 5.5 on Monday, its second model in the Claude 5.5 family after Opus 5.5 shipped a few days earlier. The company says it runs more than 30 percent faster than Sonnet 5 and costs up to 30 percent less for most work, while keeping the same per-token price of $2 for input and $10 for output.

On Terminal-Bench 4.0, a test of agentic coding tasks, Sonnet 5.5 scored 70.6 percent against Sonnet 5’s 10.3 percent. On GDPval-AA, a benchmark of real-world work across a range of occupations, it scored 1,844 against Sonnet 5’s 1,449, a couple of points behind the larger Opus 5.5 on the same test. Anthropic also reported 55.5 percent on CursorBench 4.0, a coding-assistant benchmark, and 80.1 percent partial credit on OSWorld 2.1, which tests an agent’s ability to operate a real computer desktop.

The new model ships with cybersecurity safeguards Anthropic says match Opus 5.5, including classifiers meant to block attempts to extract the model’s reasoning through distillation. It’s available now on Claude.ai for web, iOS, and Android, and through Anthropic’s API, Amazon Web Services, Google Cloud, and Microsoft Foundry.

A record that outlived four decades

At the Asian Games in Aichi-Nagoya, Vithya Ramraj ran the women’s 400m hurdles final in 54.75 seconds on Monday, a personal best that also broke the Indian national record of 55.42 seconds set by PT Usha at the 1984 Los Angeles Olympics. The mark had stood for 42 years. Ramraj’s run earned her the bronze medal. UAE’s Mariam Kareem took gold in a Games record 54.31 seconds, with Kazakhstan’s Adelina Zems taking silver in 54.58.

It was India’s best single day of these Games: eight medals in all, four silver and four bronze. In the men’s javelin final, Yashvir Singh threw a personal best of 85.72 meters on his last attempt to take silver, with Rohit Yadav improving through the rounds to 84.35 meters for bronze. Sri Lanka’s Rumesh Tharanga Pathirage won gold at 88.55 meters. It marked the second straight edition of the Asian Games with two Indian throwers on the javelin podium.

Murali Sreeshankar won silver in the long jump with a leap of 8.07 meters despite an ankle injury that forced him out of his final attempts. Gulveer Singh took silver in the men’s 1500m, clocking 3:36.72. In shooting, Esha Singh tied the Asian Games record of 38 hits out of 50 shots in the women’s 25m pistol final, forcing a shoot-off against China’s Xiao Jiaruixuan. Both hit four of five targets in the first round of the tiebreak before Xiao won the second 3-2, leaving Esha with her second consecutive Asian Games silver in the event.