Coldcard hardware wallet losses may approach $114 million, with a possible fourth sweep emerging in the ongoing attack. The incident follows an earlier report of $70 million lost from nearly 1,200 Bitcoin wallets. This matters because it indicates an ongoing, targeted attack on a widely used Bitcoin-only hardware wallet, potentially undermining user trust in hardware wallets. The scale of losses could have significant financial and reputational repercussions for the crypto community. The attack exploited weak seed generation, allowing the attacker to recreate private keys offline without ever touching the devices. Each sweep transaction paid an identical hardcoded fee of 30.0 sat/vB and left no change outputs, indicating a systematic operation.
Coldcard is a Bitcoin-only hardware wallet made by Coinkite since 2017, featuring dual secure elements and air-gapped signing. Hardware wallets are designed to keep private keys offline, but if seed generation is weak, an attacker can brute-force the keys offline. An earlier report detailed how Bitcoin cold wallets lost $70 million without the devices being touched, and the latest estimates suggest the attack is ongoing.
Anthropic has published a technical result showing that approximating the Closest Vector Problem (CVP) within a polynomial factor is NP-hard. The paper is hosted on Anthropic's official CDN and marks a significant contribution to computational complexity theory. This result strengthens the theoretical foundations of lattice-based cryptography, which underlies many post-quantum cryptographic standards. It also addresses a long-standing open question about the fundamental limits of approximating one of the most important problems in high-dimensional geometry. The Closest Vector Problem asks for the lattice point nearest to a given target in n-dimensional space. While CVP is known to be NP-hard to approximate within constant factors, this new work reportedly extends the hardness to polynomial approximation factors, a considerably stronger regime.
A lattice is a regular, infinite grid of points in n-dimensional space, and the Closest Vector Problem is a core computational problem defined on lattices. Lattice-based cryptography relies on the presumed hardness of such problems to build schemes like the NIST-standardized post-quantum algorithms. Approximation hardness results are crucial because cryptographic reductions typically require solving lattice problems approximately rather than exactly.
BlackRock is expanding its blockchain-based tokenized cash products by adding new money market funds, according to a CoinDesk report dated August 3, 2026. The move deepens the asset manager's push into on-chain finance and tokenized real-world assets. As the world's largest asset manager, BlackRock's expansion signals growing institutional demand for tokenized money market funds, which offer 24/7 trading, instant settlement, and on-chain transparency. This could accelerate the integration of traditional finance with blockchain rails and pressure competitors to launch similar products. The report did not disclose the names, sizes, or launch dates of the new funds. The one-line summary indicates the new offerings are money market funds; specifics such as additional blockchain networks or distribution partners were not included in the provided content.
Tokenized cash refers to digital representations of fiat currency or cash-equivalent assets, such as money market fund shares, issued on a blockchain. Unlike stablecoins, tokenized deposits and tokenized money market funds can preserve yield while remaining available for settlement, as noted by ABA Banking Journal. The Bank for International Settlements has called tokenization the next critical evolution of the monetary system, similar to the move from physical coins to electronic ledgers.
Binance announced it will delist and cease trading on all spot trading pairs for ACX, HFT, PIVX, PYR, VANRY, and VIC on 2026-08-17 at 03:00 UTC. The affected tokens include Across Protocol, Hashflow, PIVX, Vulcan Forged PYR, Vanar, and Viction. This delisting will remove liquidity and trading access for these tokens on the world's largest exchange, affecting holders and traders who must act before the deadline. It also signals a potential shift in Binance's listing standards, which could impact these projects' market confidence and future viability. The delisting applies to all spot trading pairs for the six tokens, with trading ceasing at 2026-08-17 03:00 UTC. Users are advised to withdraw or convert their holdings before the deadline; the exact date is more than a year away, giving ample time for action.
Across Protocol (ACX) is a cross-chain bridge protocol that allows token transfers between layer 2 networks and mainnet, with over $37 billion moved. Hashflow (HFT) is a decentralized exchange that uses a request-for-quote (RFQ) model to provide zero-slippage trades and MEV protection. PIVX is a privacy-focused, proof-of-stake blockchain that originated as a fork of Dash and emphasizes financial data protection through advanced cryptography.
OpenAI has announced GPT-Live, a new generation of voice models built in just six months that enables continuous, natural voice conversations with AI. The system uses a turnless speech model and a low-latency architecture to allow real-time interaction. GPT-Live represents a shift from turn-based voice assistants to fluid, real-time conversations, which could significantly improve user experience in applications like customer service, education, and accessibility tools. This technological leap is likely to raise the bar for voice AI across the industry, pushing competitors to adopt similar full-duplex architectures. The new model is said to have a full-duplex architecture, allowing it to listen and speak simultaneously, unlike previous turn-based systems. OpenAI has also opened a sign-up form for developers interested in accessing GPT-Live-1 through the API, indicating planned developer availability.
Traditional voice AI systems operate in a turn-based manner: the user speaks, the system processes, and then responds, which introduces noticeable delays and awkward pauses. GPT-Live's turnless design removes this back-and-forth structure, allowing for simultaneous listening and speaking, similar to human conversation. Real-time voice AI architecture typically involves connecting speech recognition, language understanding, and speech synthesis under strict latency and reliability budgets, which GPT-Live reportedly optimizes for natural interaction.
Anthropic has released a new proof establishing a super-cubic lower bound on the formula size of the permanent, a central problem in algebraic complexity. The result shows that computing the permanent requires formulas whose size grows faster than any cubic polynomial in the matrix dimension. The permanent is believed to be far harder to compute than the determinant, yet proving strong lower bounds for it has been notoriously difficult. A super-cubic formula lower bound is a rare structural breakthrough that could inform broader circuit-complexity research, and it is notable that an AI company produced results in pure theoretical computer science. In algebraic complexity, a formula is a circuit whose computation graph is a tree, and its size is usually measured by the number of leaf variables or arithmetic operations. The new super-cubic bound means the required formula size grows faster than n^3, typically expressed as Ω(n^{3+ε}) for some ε>0, strengthening earlier cubic lower bounds that came from communication-complexity and composition techniques.
The permanent of an n×n matrix is a polynomial that resembles the determinant but treats all products as positive, so it lacks the sign cancellations that make the determinant easy to compute. While the determinant can be computed in polynomial time by Gaussian elimination, Valiant's theorem shows that computing the permanent is #P-hard, and proving strong lower bounds for it is a central open problem in algebraic complexity. Formula size lower bounds, which constrain computations to tree-shaped circuits, are especially challenging because they disallow many of the sharing tricks used in general circuits.
Bitget has announced it will exit the Japanese market and close all remaining positions by the end of the year. The announcement means Japanese users will have to wind down their trading portfolios and withdraw funds before the deadline. This exit is significant because a major global exchange is leaving a strictly regulated market like Japan, which directly affects Japanese customers who must close positions and move funds. It also highlights the regulatory and operational pressures that crypto exchanges face when operating in different jurisdictions. Bitget, a centralized cryptocurrency exchange founded in 2018, will require all Japanese customers to close their remaining positions by the year-end. The company has not yet disclosed the specific reasons for its departure from Japan.
Bitget is a centralized cryptocurrency exchange, established in 2018. Japan is a significant regulated market for digital asset trading, and an exchange exiting such a market typically involves winding down operations, closing positions, and returning assets to customers. The announcement means Japanese users must act before the year-end deadline to avoid forced closure of their positions.
Bithumb, a leading South Korean cryptocurrency exchange, announced plans to pursue an initial public offering (IPO) by 2028 following a major internal restructuring. The exchange has not yet disclosed specific financial details or the listing venue. This is significant because a major crypto exchange targeting a public listing could enhance institutional adoption and improve the credibility of the cryptocurrency market. It also reflects a broader trend of crypto companies seeking access to traditional capital markets. The plan is forward-looking, targeting 2028, and comes after internal restructuring aimed at preparing the exchange for a public listing. Bithumb was founded in 2014 and is one of South Korea's largest crypto exchanges.
Bithumb is a South Korean cryptocurrency exchange headquartered in Seoul, founded in 2014 under the name Xcoin and rebranded to Bithumb in 2015. It is a centralized exchange where users can trade major cryptocurrencies like Bitcoin, Ethereum, and Ripple. Cryptocurrency exchanges pursue IPOs to access public capital markets, increase transparency, and gain regulatory legitimacy. Bithumb's 2028 IPO target follows a period of internal restructuring, reflecting its strategic efforts to prepare for a public listing.