全球领先的人工智能研究机构OpenAI近日宣布了一项重要举措——安全Bug Bounty（简称Bounty）程序。

目标与意义

该计划的主要目的是识别和防止人工智能系统中的滥用行为和安全隐患。这包括但不限于：
- 代理漏洞: AI系统可能被设计成具有自我学习能力的“代理人”，这类漏洞可能导致控制权转移。

- 提示注入: AI模型可能会对输入的文本进行分析，如果恶意用户能将特定内容作为提示输入，就可能操纵AI行为。

- 数据泄露: AI系统处理大量敏感信息时，若防护措施不足，就会面临数据被窃取的风险。

Yupp, a startup that offered a crowdsourced AI model-picking service, has announced its closure. Co-founders Pankaj Gupta and Gilad Mishne made the announcement on Tuesday.

Yupp had raised a significant seed round of $33 million in 2024, led by venture capital firm a16z, known for its investments in cryptocurrency startups. This was a major fundraising event for the company.

In addition to the large seed round, Yupp also secured checks from over 45 angels and small investors, including notable figures like Google DeepMind's Jeff Dean and Twitter co-founder Biz Stone.

However, despite these promising fundraising efforts and the involvement of high-profile individuals, Yupp was unable to sustain its business and has decided to shut down.

AI tech giant OpenAI has announced impressive revenue growth, claiming to be expanding its revenue four times faster than the companies that defined the internet and mobile eras.

The company, known for its cutting-edge AI models like GPT-5, has over 900 million weekly active users in consumer AI and more than 50 million subscribers. Search usage on their platform has nearly tripled in the last year.

OpenAI's ad pilot is already generating over $100 million in annual recurring revenue in under six weeks, suggesting a significant potential revenue stream for the company.

The company's focus on building its public market narrative is evident as it positions itself as an

Cloud software giant Salesforce has been updating its business around artificial intelligence (AI), and in a small gathering in San Francisco on Tuesday, the company's CEO Marc Benioff and his team unveiled the latest results of those efforts: an updated version of messaging platform Slack, packed with a plethora of new AI features.

The most significant among these is a serious upgrade for Slackbot, the AI-powered chatbot that serves as the primary interface for many Slack users. With 30 new AI-driven capabilities, Slackbot now has the ability to perform an impressive range of tasks, from scheduling meetings and sending reminders to transcribing audio recordings and summarizing lengthy documents.

Benioff acknowledged during his keynote address at the event that the past five years since Salesforce acquired Slack had been a

AI recruiting startup Mercor has confirmed that it was hit by a cyberattack tied to the compromise of the open source project LiteLLM. The attack, which is believed to have originated from a hacking group called TeamPCP, led to an security incident that Mercor moved promptly to contain and remediate.

LiteLLM, a widely used library for machine learning, had previously experienced a malicious code discovery in a package associated with the project. While the code was quickly identified and removed, it drew attention due to LiteLLM’s popularity on the internet.

Mercor, which works with companies like OpenAI and Anthropic to train AI models through domain experts, has not disclosed any details about data exposure or if any customer or contractor data was affected. Further investigations are ongoing.

San Francisco is about to heat up as StrictlyVC, the leading startup event platform, kicks off the new year with a bang. On April 30, at the Sentro Filipino Cultural Center in San Francisco, industry leaders, VCs, and tech enthusiasts will gather for unfiltered fireside chats, insider VC insights, and high-value connections that can make a real difference.

Anthropic's recent move to yank its leaked source code from thousands of GitHub repos has been a major headline. The company, however, clarifies it was an accident. This incident further cements Delve's troubled reputation.

As if the startup scene in San Francisco wasn't already buzzing with excitement, Anthropic seems to be having its month. The company's recent actions and developments are sure to generate more discussions among industry players.

On a more positive note, Google is now allowing US users to change their Gmail addresses. This update could potentially streamline user management for the tech giant.

Lastly, Allbirds, the sustainable shoe brand, has been acquired by a private equity firm for $39 million. Its impressive fundraising in its IPO and subsequent sale demonstrate the growing interest in eco-friendly businesses.

2026年4月30日，StrictlyVC在旧金山开启新一年的活动。这个场合为行业领导者提供了无过滤的火炉边聊天的机会，分享内部VC见解，并建立真正能推动业务发展的联系。然而，这场活动的一个意外插曲却让整个科技新闻界都震动了。

软件工程师在周二发现，Anthropic——一家以其领先的人工智能产品而闻名的公司——在最近的一次发布中，无意间包含了访问其最受欢迎产品Claude Code命令行应用源代码的链接。AI爱好者们仔细研究了泄露的代码，希望能从中窥探到Anthropic如何利用在其应用底层的大型语言模型（LLM）来提升产品的性能。

这一意外事件引发了GitHub上的大规模行动。Anthropic收到了美国数字版权法下的删除通知要求GitHub移除包含违规源码的仓库。据GitHub记录显示，这个通知被执行到了大约8,100个代码库上——其中包括一些合法的Anthropic Claude Code分支。

愤怒的社交媒体用户和受到牵连的开发者们在事件中发出了强烈的抗议声音。Anthropic的Claude Code负责人Boris Cherny随后发表声明，称这是一个意外，并撤回了大部分删除通知，只保留了一个受影响的仓库以及96个附带的fork分支。

这次GitHub上的混乱事件无疑给 Anthropic 的声誉带来了新的挑战。而随着公司计划进行首次公开募股（IPO），这样的负面新闻可能会对其上市进程产生影响。然而，对于科技行业来说，每一次危机都是一次学习和改进的机会。让我们拭目以待Anthropic如何从中吸取教训，并在未来更好地服务其用户和技术社区。

Google has recently updated its policies, allowing US-based Gmail users to modify their email addresses. This move is part of Google's ongoing efforts to improve user experience and provide more flexibility in managing online accounts.

Microsoft's AI research lab has released three new foundational AI models. These models aim to generate text, voice, and images, showcasing the tech giant's continued push in multimodal AI.

Model Details

• MAI-Transcribe-1: This model transcribes speech across 25 different languages into text. According to Microsoft, it is 2.5 times faster than Azure Fast offering.

• MAI-Voice-1: A voice-generating model that allows users to create custom voices in seconds.

• MAI-Image-2: A video-producing model capable of generating high-quality videos from input text or audio.

Relevance and Impact

These models mark a significant step forward for Microsoft's AI capabilities. By offering more cost-effective alternatives to rival models, such as those from Google and OpenAI, Microsoft aims to position itself at the forefront of multimodal AI research.

The release of these foundation models is likely to stimulate further development in the field, potentially leading to even more advanced AI applications.

OpenAI, the AI research lab on the brink of an IPO, has made its first acquisition in the media industry. The company has acquired popular tech talk show TBPN.

TBPN, hosted by former tech founders John Coogan and Jordi Hays, is a daily live show that airs on YouTube and other platforms. The show focuses on topics such as technology, business, AI, and defense.

Acquiring TBPN aligns with OpenAI's mission to promote the responsible development and use of AI. By acquiring a popular media platform, OpenAI aims to expand its reach and influence in shaping public discourse around AI.

The acquisition will likely involve integrating TBPN's content and audience into OpenAI's existing platforms and initiatives. This move is expected to strengthen both companies' positions in the growing AI ecosystem.

AI chatbots and generated imagery have brought increased legal and reputational pressure on AI companies. To address these failures in safety guardrails, Moonbounce has introduced a novel approach to policy management.

Policy as Code

Moonbounce's 'policy as code' principle involves turning static policy documents into executable, updatable logic tightly coupled with enforcement mechanisms.

How it Works

This process allows Moonbounce's system to dynamically evaluate content at runtime, provide real-time responses, and take appropriate actions based on customer preferences. These actions could range from slowing down distribution to blocking high-risk content.

Moonbounce in Action

Today, Moonbounce supports various verticals where AI-generated content is prevalent. Some of these include platforms dealing with user-generated content like dating apps, AI companies building characters or companions, and AI image generators.

By providing a robust safety layer, Moonbounce empowers its customers to not only improve their products' safety but also differentiate themselves in the market by highlighting their commitment to user well-being.

As artificial intelligence (AI) continues to demand exponential amounts of power, tech companies are racing to secure natural gas supplies for their data centers.

The US has abundant natural gas reserves, but shipping the fuel isn't cheap. This has led to a certain level of insulation for the country from Middle East turmoil that affects oil prices and gas supplies.

However, the growth in production in the 'big three' regions - responsible for three-quarters of all US shale gas production - has slowed significantly lately. The future of these supplies is uncertain.

The reliance on natural gas for AI data centers is significant, as it accounts for about 40% of electricity consumption in the US, according to the Energy Information Administration. This means that fluctuations in natural gas prices directly impact electricity costs and ultimately the operations of data centers.

Tech companies are trying to move their gas power plants behind the meter - bypassing the grid and connecting directly to their data centers. However, this strategy relies on a finite resource, which could become a limiting factor for these companies in the long run.

The AI boom has highlighted the physical constraints of the digital world. Tech companies might regret chasing the FOMO (Fear Of Missing Out) by betting big on a finite resource that could prove to be a volatile and ultimately constraining factor.