Breaking

Meta Unveils Low-Cost 'Muse Spark 1.1' AI Model Trained on Employee-Tracking Data

July 9, 2026 at 12:50 EDT

Foundation Models
AI Agents
Policy & Regulation

On July 9, 2026, Meta released Muse Spark 1.1, a low-cost AI model specialized in agentic and coding tasks. Available through the newly launched Meta Model API and Meta AI, the model is drawing attention because its training used data from an internal program that recorded US employees' computer activity.

July 9, 2026 · Meta Superintelligence Labs

Meta's Muse Spark 1.1 undercuts rivals on price — but its data came from watching employees

A low-cost agentic coding model built to operate desktop, mobile, and browser interfaces — priced at roughly a quarter of competitors. Its training drew on an internal program that recorded US employees' keystrokes, clicks, and screen snapshots.

~1/4

of competitors' pricing — its main weapon

token context window

1,600

employees signed a petition against the data program

Benchmark scores

Higher is better · scores out of 100

88.1

MCP Atlas

80.0

Terminal-Bench 2.1

62.1

HLE (with tools)

61.5

SWE-Bench Pro

The controversy — Model Capability Initiative

An internal program recorded US employees' computer activity to train AI agents — then it was paused amid privacy backlash.

WHAT WAS CAPTURED

Keystrokes, mouse movements & clicks, periodic screen snapshots — across Gmail, GitHub, Slack, Wikipedia.

WHY IT STOPPED

Paused around June after collected data became broadly accessible internally — exposing private chats, performance, tax and medical records.

SEEN AS STRENGTHS

Pricing at ~1/4 of Anthropic and OpenAI rivals
Distribution across a vast existing user base
Parallel sub-agents & native desktop/mobile/browser use

SEEN AS RISKS

Training data from employee surveillance
Sensitive records exposed internally
Viewed by some as tracking staff to replace them with AI

Planned rollout

WhatsApp Instagram Facebook Messenger AI glasses

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.