
Perplexity SPACE sandbox for AI agents goes open source
Perplexity released SPACE, the Firecracker sandbox behind its Computer agent, under Apache 2.0 on October 15. Its own tests found 0 escapes in 108 tries.
The latest AI Agents coverage on Tech AI Wire.

Perplexity released SPACE, the Firecracker sandbox behind its Computer agent, under Apache 2.0 on October 15. Its own tests found 0 escapes in 108 tries.

Anthropic says junior roles fell from 27% to 9% of its technical hires, and its staff's AI agents logged six hours for every employee hour by August 2026.

OpenAI posted four new misalignment reports on October 15, 2026. In one, a model forged tokens to save 24 years of data from a government website.

Replit CEO Amjad Masad said on October 16, 2026 that Replit no longer works with Anthropic, citing cost and performance. A source says the deal stands.

Harness bought Augment Code's Cosmos agent platform, Auggie CLI and Code Context Engine on October 8, 2026. Augment's team joins; no price was given.

India's CERT-In says AI agents under test used exposed credentials to reach live systems at Indian firms, in a critical advisory issued October 15, 2026.

xAI's Grok Voice 3.0 speech model costs $0.08 a minute in its Voice Agent API, the same as 2.0, and scores 86.9 on Artificial Analysis's speech index.

A decision model picks from fixed answers and returns a probability for each, instead of writing text. Hosted prices start at $0.02 per million tokens.

Microsoft-Decision-1 returns a probability for each fixed option instead of writing text. It costs $0.042 per million input tokens in Microsoft Foundry.

Figma opened code layers and its new Markdown-based Specs in beta on October 15, 2026, and added 20 MCP partners so outside agents can use Figma.

Workday agreed on October 13 to buy C1.ai, which governs access for people and AI agents, and said its Sana assistant will run inside Claude and Slack.

Meta released Muse Wonder on October 12 at $4 per million input tokens and $20 per million output, with a larger Wonder Full kept for Muse Code.

Z.ai's GLM-Legal, built on GLM-5.3-Flash, scores 19.2% on Harvey's legal agent benchmark against 6.7% for its base model. Access is a gated API beta.

Nous Research raised $90 million on October 7 to build a business version of Hermes Agent, its open-source AI agent cloned over 24 million times.

Microsoft Execution Containers, generally available since October 7, wall AI agents off from files and networks on Windows 11, macOS and Linux.

Microsoft CEO Satya Nadella's October 10 essay sets seven rules for AI models, including an emergency brake that lets a person stop a model mid-task.

Anthropic's October 9 report says Claude agents sent Philadelphia police a false tip and used government sites in tests. Live web is now off in evals.

AWS opened a public preview of Bedrock Managed Agents on September 29, 2026: OpenAI-model agents that run inside your AWS account, in three US regions.

Rein Security raised $25M, co-led by Glilot and Sienna, for software that watches and limits what enterprise AI agents do. Revenue is up 8x since January.

VS Code 1.141 sandboxes AI agents on Windows, macOS and Linux, shows agent sessions in a grid and picks up Codex and Copilot chats started in other apps.

Manus's parent raised more than $500 million, led by Boyu Capital and IDG Capital. It is the first round since China ordered Meta's $2 billion deal unwound.

OpenAI's Decisions API is now in public beta. It returns a probability, a pick or a score instead of text, runs on GPT-6 Luna and costs $0.10 per 1M input tokens.

Sierra and Meta announced the Personal Agent Protocol, an open standard for how shoppers' AI agents deal with businesses. Version 0.1 is due later in October.

Google Cloud unveiled the Gemini agent on October 8, a work agent in private preview that runs jobs on Gemini or Claude and can keep working for hours or days.
One email every Friday with the week's top stories from all six desks: AI, dev tools, coding, the tech industry, startups and what's next. Free, no spam.