To mitigate AI risks, companies should be transparent about their systems, collaborate with the industry, stress test, and share details about their progress
The case for transparency is growing as the best way to combat fears on the developing technology — The writer is president of global affairs at Meta
Context & Ripple Effects
Meta's global affairs president is making the voluntary-transparency case just as the argument over who defines AI safety has turned personal inside the company itself: earlier this year Meta's Yann LeCun accused DeepMind's Hassabis, Altman, and Amodei of fearmongering to achieve regulatory capture. The op-ed reads as Meta staking out the opposite pole — openness and stress testing instead of alarm.
It also arrives against an awkward internal backdrop. Meta was among the big tech firms that cut their responsible AI teams in 2023, so the pitch for industry collaboration and shared progress doubles as an attempt to rebuild governance credibility by process rather than headcount. The demand-side signal is already there: a recent study found roughly 75% of S&P 500 firms have expanded their AI risk-factor disclosures.
First-order effects
- Meta positions itself as the governance-minded alternative to the safety-alarm camp, forcing rivals like DeepMind and Anthropic to answer either with matching disclosure commitments or by defending their stricter stance.
- Corporate buyers get a template — transparency, cross-industry collaboration, model stress tests, shared progress reports — that they can demand from vendors as evidence of diligence.
Second-order effects
- If Meta's framework gains traction, disclosure shifts from differentiator to baseline: firms that already updated risk disclosures under investor pressure face a second round, this time covering system behavior rather than legal exposure.
- Voluntary stress testing preempts part of the regulatory agenda — giving policymakers an industry-built audit format to adopt, and weakening the argument that only mandatory regimes can produce usable safety data.
Third-order effects
- The endgame the corpus points toward is convergence between Meta's voluntary sharing and Amodei's push for mandatory third-party testing of cyber, bio, and autonomy risks — self-disclosure hardening into audited standards if legislators borrow the industry's own format.
- Governance capacity thus becomes structural: labs that can demonstrate external verification will be the ones regulators deem safe to operate frontier systems, splitting the industry into assurable and unassurable players.
The trend: Frontier labs are racing to define AI safety through voluntary disclosure and stress testing before regulators convert those practices into mandatory audit regimes.