California Safety Bill, OpenAI and Anthropic x US AI Safety Institute, Meta Apologizes, Telegram Co-founder Arrested, Google 10th Zero-Day Attack

GenAI Safety & Security Newsletter (Aug 26 - Sept 2, 2024)

[FREE e-book] GenAI safety and security

Highlights

1. California Passes AI Safety Bill: California legislators passed SB 1047, mandating safety testing, disclosure of compliance, and legal actions against misuse.

2. OpenAI, Anthropic Partner with US AI Safety Institute: Both agreed to test their new AI models prior to public release, focusing on mitigating potential risks.

3. Meta Admits to Political Censorship: Mark Zuckerberg confessed to bowing to pressure to censor certain COVID-19 expressing regret for the decision.

4. Telegram Founder Arrested in France: Pavel Durov, co-founder of Telegram, faced arrest and formal investigation in France for illegal activities.

5. Google Hit with 10th Zero-Day Attack: North Korean hackers exploited a zero-day vulnerability in Google Chrome to steal cryptocurrency.

Deep Dive

1. California AI Bill Passes Legislature, Awaits Governor's Signature:

• The bill, SB 1047, mandates safety testing for advanced AI models costing over $100 million to develop and provides the state attorney general with the power to sue non-compliant developers. Reuters

• Requires developers to hire third-party auditors to assess their safety practices and provide additional protections to whistleblowers speaking out against AI abuses.Reuters

• This landmark legislation could set a precedent for states' AI regulation across the U.S. after the Biden Executive Order on Artificial Intelligence in Oct, 2023

Next: Gov. Gavin Newsom will decide to sign it into law or veto it in September.

2. OpenAI and Anthropic Partner with US AI Safety Institute:

• The agreement allows the institute, established under President Biden's executive order on AI safety, to access new AI models from both companies before and after public release. CNBC

• The collaboration aims to increase research on the capabilities and risks of AI, evaluate safety risks, and develop methods for mitigating those risks. NIST

• OpenAI CEO Sam Altman and Anthropic co-founder Jack Clark have both expressed support for the agreement and the U.S. AI Safety Institute's mission. The Hill

Comment: Given the intense ongoing cyber attacks, there will be more pushes to ensure public AI safety to avoid "mistakes" with social media.

California's AI bill, Harris x AI safety, IPv6 "wormfood", Quantum Cryptography Standards, Meta Data Crawler

3. Meta Apologizes for COVID-19 Censorship

• Meta CEO Mark Zuckerberg admitted to succumbing to pressure from the Biden administration to censor content on Facebook and Instagram related to COVID-19, including humor and satire, during the pandemic.

• He expressed regret. He believes the government pressure was wrong and that Meta would be prepared to push back if it happened again. The Guardian

• This admission raises critical questions about the balance between government influence and free speech on social media platforms Vox

Comment: If the US can not regulate AI, goverment are likely to remove the section 230 to regulate all publishers and the platforms on the internet The Hill and Valley

4. Telegram Boss Arrested in France, Potential 10-Year Sentence: Reuters

• Pavel Durov, co-founder of the messaging app Telegram, was arrested in France.

• French authorities allege that Telegram has failed to respond adequately to requests for information regarding alleged complicity in child exploitation and organized crime on the platform known for end-to-end encryption.

• He was released after 4 days, but could face up to 10 years in prison if convicted.

Comment: A turning point in how EU gov approach holding tech executives accountable for illegal activity on their platforms after EU AI act takes effect.

5. Google Hit with 10th Zero-Day Attack This Year:

• Google patched the vulnerability, but found by Microsoft.

• A North Korean hacking group exploited a zero-day vulnerability in the Chrome browser to target cryptocurrency users and steal digital assets. TechCrunch

• The vulnerability allowed the hackers to gain remote code execution, giving them control of victims' computers and access to sensitive information. Forbes

Action: The patch is available. Please update your Chrome to version 128 or later.