Anthropic推出15000美元的漏洞赏金计划,以加强人工智能安全

币界网Pubblicato 2024-08-08Pubblicato ultima volta 2024-08-08

币界网报道:

亚马逊支持的人工智能初创公司Anthropic推出了一项漏洞赏金计划,并将为每份发现其人工智能系统关键弱点的报告支付高达15000美元。该倡议是任何使用高级语言模型众包安全测试的公司所做的最广泛的努力之一。

据该公司称,赏金针对的是“普遍越狱”攻击,这种方法可以绕过生物武器和网络威胁等领域的人工智能安全措施。在向公众提供下一代安全缓解系统之前,Anthropic计划允许道德黑客对其进行测试,以防止潜在的滥用。

Anthropic的漏洞赏金计划最初是与HackerOne合作开展的一项仅限受邀者的计划,该计划希望网络安全研究人员具备识别和修复其人工智能系统漏洞的技能。该公司计划在未来更广泛地开放它,有可能提供一种全行业人工智能安全合作模式。

与此同时,英国竞争与市场管理局(CMA)正在调查亚马逊对Anthropic 40亿美元的潜在竞争问题投资。在监管审查日益严格的背景下,关注安全可以提高Anthropic的声誉,使其与竞争对手区别开来。

Anthropic制定了新的AI安全标准

虽然OpenAI和谷歌也有漏洞赏金计划,但它们主要关注传统的软件漏洞,而不是人工智能特有的漏洞。Meta因采取了一些人认为相对封闭的研究方法来确保越来越智能的机器的安全开发而受到批评。通过明确针对这些问题并邀请外部对其进行审查,Anthropic为该行业的开放树立了先例。

然而,人们怀疑仅靠漏洞赏金能否有效解决与保护先进机器学习系统相关的所有问题。虽然对于识别和修补特定缺陷很有价值,但它们可能无法应对围绕人工智能对齐和长期安全的更广泛挑战。可能需要一个更全面的战略,包括广泛的测试、改进的可解释性和潜在的新治理结构,以确保人工智能系统在变得更加强大时与人类价值观保持一致。

Letture associate

The $6.4 Billion CRO Accumulation Plan of Trump Media Group Falls Apart

A $6.4 billion plan to create a publicly-listed CRO treasury company, announced a year ago by Trump Media & Technology Group (DJT), Crypto.com, and SPAC Yorkville, has been terminated. The ambitious deal, which aimed to accumulate approximately 6.3 billion CRO tokens (nearly 20% of supply at the time), never progressed beyond a framework agreement. Related plans for a prediction market integrated into Truth Social and ETF custody services by Crypto.com were also shelved, scaled back to a simple marketing partnership. The collaboration followed significant political alignment, with Crypto.com donating to Trump's inauguration and a pro-Trump super PAC, and its CEO meeting with Trump. The SEC also closed an investigation into the exchange shortly before the deal was announced, raising conflict-of-interest concerns. Officially, the termination was attributed to an unfavorable market environment. CRO's price has fallen roughly 70% since the announcement, and the broader market for publicly-traded digital asset treasuries has cooled significantly, with Bitcoin nearly halving from its late-2025 peak. The only completed transaction from the 2025 agreements remains intact: Trump Media's ~$105 million CRO purchase and Crypto.com's $50 million purchase of DJT stock. The termination aligns with DJT's strategic pivot away from crypto; the company is now pursuing a multi-billion dollar all-stock merger with nuclear fusion firm TAE Technologies, shifting its focus to clean energy.

marsbit13 min fa

The $6.4 Billion CRO Accumulation Plan of Trump Media Group Falls Apart

marsbit13 min fa

Tiger Research: The $43 Million Gray Area of Asian Prediction Markets

Tiger Research: Asia's $43 Million Grey Zone in Prediction Markets Western jurisdictions have created regulatory pathways for prediction markets through derivatives law (like the U.S. CFTC framework) or flexible gambling licenses (like the UK's "betting intermediary" category). In contrast, Asia lacks a comparable regulatory architecture. The absence of a general gambling license framework adaptable to private operators, coupled with closed financial product definitions (e.g., positive lists of assets in Korea and Japan), leaves prediction markets in a regulatory grey area. This lack of a clear classification—whether as gambling, financial derivatives, or a novel third category—has not stopped market activity. Significant liquidity, evidenced by over $52 million tied to a single South Korean election, flows to offshore platforms. However, this results in forgone tax revenue (estimated at $4-43 million annually per major Asian market), no consumer protection, and no oversight for market integrity. The core issue is not cultural opposition to gambling, which exists legally in many Asian markets, but a missing institutional design. Without a deliberate regulatory pathway, authorities are left with suboptimal options: expanding criminal sanctions or blocking access, which fail to address the underlying economic activity or achieve key policy goals. Establishing a regulatory foundation requires initiating a formal public discourse to first define prediction markets' legal status and societal value, a foundational step that has yet to occur in most Asian jurisdictions.

marsbit23 min fa

Tiger Research: The $43 Million Gray Area of Asian Prediction Markets

marsbit23 min fa

How Singapore's Retail Giant Masters the Gold Trade: Selling Thousands of Pounds Monthly Without Betting on Price, Gold Leasing Inspires RWA Income Design on Chain

A Singapore retail giant, Mustafa Centre, sells around 1,100 pounds of gold jewelry monthly, holding nearly a ton of inventory valued over $100 million, yet avoids gold price speculation. They achieve this by maintaining a constant inventory—buying back the exact amount of gold sold each day, earning only from sales margins, not price movements. This practice is enabled by the century-old gold leasing market. Retailers borrow physical gold to meet inventory needs without tying up capital or bearing price risk, paying a lease rate. Lenders earn yield on otherwise idle gold. This real-economy demand, driven by operational needs rather than speculation, provides a stable income stream. The article highlights how this traditional leasing model inspires on-chain Real World Asset (RWA) protocols like thUSD and thGOLD. These protocols channel gold lease yields—generated from borrowers like Mustafa—to token holders. The key challenge has been access, not yield availability. Platforms like Libeara, with institutional backing (e.g., SC Ventures), facilitate this by tokenizing funds (e.g., the MG 999 Gold Fund) that lend to vetted commercial borrowers, making this off-chain yield accessible on-chain. Thus, gold leasing demonstrates a proven, non-speculative revenue source for crypto protocols seeking sustainable "real yield," connecting physical trade finance to decentralized finance.

marsbit33 min fa

How Singapore's Retail Giant Masters the Gold Trade: Selling Thousands of Pounds Monthly Without Betting on Price, Gold Leasing Inspires RWA Income Design on Chain

marsbit33 min fa

Trading

Spot
活动图片