Meta's AI agent Muse draws privacy scrutiny as Anthropic bets big on training
Meta's AI agent Muse is under fire for building detailed profiles of users' contacts, while retailers split on adopting it. Anthropic commits $100 million to train 10,000 workers, and an ex-Anthropic security lead warns AI agents are outpacing human oversight.
Every line links to a primary source in the full briefing.
01Meta's Muse quietly profiles your contacts, WIRED investigation finds
WIRED reports that Meta's AI agent Muse, downloaded by millions, builds detailed profiles of users' friends and family in order to carry out tasks on their behalf. The investigation says this comes with real privacy costs that are not obvious to users when they first set up the assistant.
If your business or staff use Muse for scheduling, messaging or customer contact, assume it is collecting and retaining data on everyone in those conversations, which raises data protection questions for any EU-based company.
Source: WIRED
02Shopify and Amazon split on whether to plug Muse into their platforms
Shopify and Amazon have reportedly taken opposite positions on integrating Meta's AI agent Muse into their e-commerce ecosystems. The divergence suggests retailers are still working out whether the benefits of a third-party AI agent outweigh the risks of ceding customer interaction to it.
If you sell through Shopify or Amazon, watch which platform you are on, since one may soon offer deeper AI agent integration for customer service or product discovery while the other holds back.
Source: finance.biggo.com
03Meta opens up AI hardware with a DIY Muse kit, and makes its app mandatory for glasses
Meta launched Muse Gadgets, an open-source project letting hobbyists build their own AI hardware on ESP32 boards and connect it to Muse, alongside a small production run of a USB-C smart home device. Separately, Meta's AI companion app, now required for using its smart glasses, added email and calendar access in its latest update.
Meta is positioning Muse as the control layer for a growing range of hardware and personal data access, which is worth tracking if you or your team use Meta's glasses or smart home products for work.
Source: The DecoderMIXED Reality News
04Former Anthropic security lead says AI agents are outrunning human oversight
A former security leader at Anthropic publicly warned that AI agents are becoming too autonomous for people to reliably monitor and control. The warning comes from someone with direct insider experience building these systems, not an outside critic.
If you are deploying AI agents to handle tasks like customer replies or data processing, build in human checkpoints now rather than assuming the agent will flag its own mistakes.
Source: 930 WFMD Free Talk
05Anthropic puts $100 million into training 10,000 workers at McKinsey and Deloitte
Anthropic announced a $100 million commitment to train 10,000 engineers in AI skills, starting with staff at consulting firms McKinsey and Deloitte. The program aims to build a larger pool of people who can build and deploy AI systems in enterprise settings.
Expect your consultants and advisors from major firms to arrive with more AI fluency soon, which could change what they recommend and how quickly they can implement it.
Source: MIXED Reality News
06Developer community pushes for hard spending limits on AI agents and APIs
A prominent independent developer argued that pay-per-use AI services badly need default hard budget caps, meaning usage is cut off entirely once a spending limit is hit, not just flagged with a warning email. The piece points to coding agents in particular as a source of runaway, unpredictable costs.
If you or your team use AI coding agents or pay-by-usage AI APIs, check today whether your provider offers a hard cutoff, because a soft warning will not stop a bill from spiraling overnight.
Source: Simon Willison
07Eurozone firms self-fund AI investment while US rivals borrow and raise capital
A report finds that companies in the eurozone are financing their AI investments mainly out of their own funds, in contrast to American competitors who lean more on external financing such as debt and equity markets. The pattern points to a more cautious, balance-sheet-driven approach to AI spending in Europe.
If you are a European business weighing AI investment, you are in good company sticking to cash-funded, incremental spending rather than taking on debt to chase AI capability.
Source: Informat.ro
Sources
8 links- 01We're going to need default hard budget caps on pretty much everythingSimon Willison
- 02Shopify and Amazon's Divergent Choices Over Meta's AI Agent 'Muse'finance.biggo.com
- 03"Muse Gadgets" turns AI hardware into an open-source DIY projectThe Decoder
- 04Muse Creates Detailed Profiles of All Your Friends and FamilyWIRED
- 05Former Anthropic security leader warns AI agents are becoming too autonomous for humans to keep them in check930 WFMD Free Talk
- 06Las empresas de la zona euro financian sus inversiones en inteligencia artificial con fondos propios, a diferencia de sus rivales estadounidensesInformat.ro
- 07Meta's AI app is the required companion for its glasses, and 292.1.0 adds email and calendarMIXED Reality News
- 08Anthropic commits $100 million to train 10,000 engineers, starting with McKinsey and DeloitteMIXED Reality News