Zuckerberg leaked audio shows Meta used employees to train AI
Leaked audio from Meta all-hands reveals Zuckerberg justified keystroke surveillance of staff to outpace AI rivals
Meta's median employee compensation dropped from $417,400 to $388,200 last year, and while that pay cut was sinking in, the company was also logging their every keystroke. Leaked audio from an April 30 all-hands meeting, obtained by More Perfect Union, captures CEO Mark Zuckerberg explaining a mandatory surveillance programme in terms his company never used publicly.
The leaked audio claimed that its own engineers were chosen as AI training subjects precisely because they are smarter than the contractors the rest of the industry relies on.
The programme referred to internally as the Model Capability Initiative, as reported by Reuters in April, runs on an approved set of applications that includes Gmail, GChat, VSCode, and Meta's internal AI assistant Metamate. This program collects mouse clicks, keystrokes, and occasional screenshots of employees' devices.
According to Zuckerberg, the content collected through the programme contains no identifying information and is not watched by any human reviewer. "It's purely just like we are using this to feed a very large amount of content into the AI model," he said.
There were no options left for the workers. The most common remark after making the announcement was to ask if there was a way out. Meta’s CTO Andrew Bosworth responded in one sentence, saying that there is no way to opt-out if you use a company computer. The response got over a hundred emojis from shocked readers.
"The average intelligence of the people who are at this company is significantly higher than the average set of people you can get to do tasks if you're working through these contractors," he said.
The training of an AI system on the real workings of senior engineers will help Meta become a better coder faster than anything competitors can match, he claimed. This was no accident. "It is not strategically in your interest for us to communicate everything in all the detail that we normally would on this," Zuckerberg conceded, knowing well that there was bound to be a leak from the meeting.
-
Meta releases Muse Spark for Mac-optimized AI: What to know
-
OpenAI, Anthropic subpoenaed by FTC over AI safety incidents and consumer harms
-
China’s ByteDance makes major AI safety push with global hiring drive
-
Can Wall Street keep rogue AI out of its systems?
-
Tim Cook announces major Apple Intelligence update for China
-
OpenAI sued by eight states over ChatGPT youth safety and mental health risks
-
1.3GW solar farm, 1GW data centre planned near King's Lynn
-
AI agent hacks are 'early warning shots': Google DeepMind co-founder
-
Russia cracks down on AI bears as real attacks spread fear: What to know
-
UK media watchdog warns X over illegal hate content despite moderation speed improvements
-
OpenAI vs. Apple: Former employee accused of taking unreleased device files
-
Google DeepMind, Google Cloud grant European researchers access to frontier AI models and $3M in compute
