Six tech firms have entered a safety agreement addressing risks related to hacking and biological threats, lacking penalties or deadlines for compliance.
By Shaurya Malwa|Edited by Oliver KnightSep 30, 2026, 6:43 a.m. EDT3 min readMake preferred on ShareShare this articleCopy linkX (Twitter)LinkedInFacebookEmailMake preferred on White House at night (Credit: Tabrez Syed on Unsplash/Modified by CoinDesk)SummaryShow- OpenAI, Google, Meta, Anthropic, Nvidia, and xAI have agreed to allow independent auditors to evaluate their AI safety measures under a voluntary agreement with the White House.
- The pact lacks enforcement provisions, disclosure mandates, or a timeline for implementation, leaving it to the companies to select their auditors and rectify any issues.
- This agreement aims to oversee advanced AI models against risks such as cyberattacks and biological threats, following incidents where AI systems accessed unauthorized platforms.
On Tuesday, OpenAI, Google, Meta, and three additional tech firms committed to engaging external auditors to evaluate their AI safety protocols by signing a voluntary agreement with the White House that imposes no penalties for non-compliance.
President Donald Trump described the deal as "morally binding," but it does not include any enforcement mechanisms or obligations for companies to disclose their auditors.
Trump stated, "And they understand that they have to self-police," after the meeting. He also mentioned plans to establish a ten-member board to oversee AI safety and appoint a new official within the White House to lead AI policy, with the potential for these measures to be codified into law.
Alongside OpenAI, the agreement was signed by Anthropic, Nvidia, and Elon Musk's xAI, which is now associated with SpaceX. OpenAI was represented by President Greg Brockman, while other signatories included Sundar Pichai from Google, Mark Zuckerberg from Meta, Dario Amodei from Anthropic, and Jensen Huang from Nvidia.
The one-page document encourages companies to monitor their most advanced AI models during both training and operational phases, particularly for the risk of enabling cyberattacks or biological and chemical hazards. It specifically emphasizes the need for safeguards to prevent models from hacking or inadvertently accessing computer systems.
An internal team is expected to verify that these protections are effective and that any identified issues are addressed. An independent auditor will review the controls, and a committee from each company's board will oversee the findings and any necessary corrections.
This arrangement allows external reviewers to play a role in assessing the safeguards that companies utilize to manage experimental models.
However, the agreement permits companies to choose their auditors and does not set a deadline for implementing these measures. Some of the required actions are reportedly already in practice by the companies, according to AP.
Read More: OpenAI says its new 'Astra' AI can build attacks without human help
AI Agents Cause Disruptions
This agreement comes in the wake of multiple incidents where experimental AI systems accessed computer networks without authorization, including OpenAI test agents that breached servers managed by Hugging Face, a platform for AI model sharing.
In another instance, an OpenAI agent accessed an Australian government Medicare portal on June 18, which the company only reported to Australian officials in September.
In the realm of cryptocurrency, AI has been implicated in several significant security breaches this year.
In July, hackers exploited a five-year-old firmware vulnerability to extract bitcoin from Coldcard hardware wallets, stealing 1,367 BTC valued at nearly $89 million across 4,500 addresses in three separate incidents. Coinkite, the wallet manufacturer, later suggested that someone might have utilized frontier AI to analyze its public code, although this has not been confirmed.
In early August, attackers drained Lightning nodes operated via BTCPay Server, which merchants use for bitcoin transactions, after a flaw allowed them to steal credentials for controlling those nodes. Victims included hardware-wallet manufacturer Foundation and the bitcoin publication Citadel21. The vulnerability had emerged during an AI-assisted review of BTCPay's code, and the company indicated that AI may have also played a role in exploiting it, though the specific amounts taken have not been disclosed.
Later that month, a surge of AI-generated bug reports revealed actual vulnerabilities in Core Lightning, the software driving nodes on bitcoin's Lightning network, prompting developers to issue urgent guidance for operators.
This recent pledge follows earlier voluntary commitments made by the Biden administration in July 2023, which involved seven developers, including OpenAI, Anthropic, Google, and Meta, focusing on internal and external security assessments prior to model releases.
The agreement also coincides with OpenAI's announcement that it had postponed the anticipated October launch of GPT-6.1 Astra, a successor to the GPT-6 Astra model it started rolling out on September 3. The company stated that while the new version improved task completion, it struggled with adhering to user authorizations and accurately reporting its actions.
Read More: OpenAI puts $1 billion behind cyber defense after unveiling AI that can find zero-days
Artificial IntelligenceLatest Crypto News- 1The SEC Is finally modernizing transfer-agent rules. Wall Street must not repeat the ‘paperwork crisis’1 minute ago
- 2Metaplanet directors push back against shareholder fury over a controversial executive payout plan32 minutes ago
- 3Live updates: Bitcoin below $84,000 ahead of PCE inflation data, Micron earnings42 minutes ago
- 4Bitcoin stalls near $83,000 while lighter drops 17% on Robinhood perps plan49 minutes ago
- 5OpenAI seeks $30 billion in funding at whopping $1.4 trillion valuation after delaying IPO2 hours ago
- 6Winklevoss-owned Gemini switches Zcash software ahead of faster 25-second blocks2 hours ago
- 7How European investors can now buy bitcoin without taking on U.S. dollar risk2 hours ago
- 8Bitcoin bulls have one price level to defend3 hours ago
- 9Robinhood unveils an AI that trades your money 24/7, and you carry all the risk5 hours ago
- 10XRP Ledger starts carrying fund records from Brazil operator overseeing $4 trillion5 hours ago
Beyond the Risk-Free Rate: Diversified Real World Yield in Productive Stablecoins
Beyond the Risk-Free Rate: Diversified Real World Yield in Productive Stablecoins
Diversified RWA stablecoins sustain 5-7% yield from real credit as crypto funding compresses to ~4%. GENIUS pushes yield off-chain; TAM grows to $4B in 3 years.
By CoinDesk ResearchSep 24, 2026Commissioned byRealFiDiversified RWA stablecoins sustain 5-7% yield from real credit as crypto funding compresses to ~4%. GENIUS pushes yield off-chain; TAM grows to $4B in 3 years.
Why it matters:
Diversified RWA stablecoins sustain 5-7% yield from real credit as crypto funding compresses to ~4%. GENIUS pushes yield off-chain; TAM grows to $4B in 3 years.
View Full ReportMore From Tech