A heavy 24-hour news cycle in crypto and adjacent tech markets was led by SpaceX’s first earnings report as a public company, which showed $7.8 billion in second-quarter revenue, 18,712 BTC on its balance sheet, and roughly $540 million in unrealized losses after Bitcoin fell 33% in the quarter. Coldcard’s wallet security incident kept escalating as the company urged users to migrate funds, while Galaxy Digital’s Alex Thorn said at least 15 attackers have now been identified through victim reports and on-chain tracing. In South Korea, Upbit and Bithumb both moved to list QUID, while on-chain flows highlighted a new wallet buying CASHCAT, a 40x leveraged BTC short on Hyperliquid, HYPE unstaking, and another Strategy-linked BTC transfer. Elsewhere, U.S. crypto policy centered on the CLARITY Act and fresh political friction around Trump’s TRUMP meme coin, while institutions expanded tokenized finance, staking, and custody services through Wells Fargo, BNY Mellon, Circle, Dinari, and Cloudflare. Markets also tracked BIP-110 activation risk, lower Bitcoin implied volatility, Ethereum staking policy debate, major AI infrastructure deals, and a long list of company, macro, and semiconductor updates.

The Linux Foundation has released a draft of the Shared AI Findings Exchange, or SAFE, a proposed set of guidelines aimed at turning cybersecurity incidents in agentic AI into shared defensive capabilities for the wider ecosystem. The draft was announced as the annual Black Hat security conference opened in Las Vegas, according to NVIDIA’s blog. The effort is being developed through the Open Secure AI Alliance, which now has more than 120 member organizations. The SAFE guidelines were drafted by a working group within the alliance, with support for the initial proposal coming from Linux Foundation collaborators including NVIDIA, Cisco Systems, CrowdStrike, Hugging Face and Red Hat. The draft lays out several recommended practices. These include collecting and analyzing AI security incidents and “near misses” in a confidential manner, notifying affected parties, identifying recurring failures in security controls, and issuing evidence-based operational guidance designed to reduce risk across the broader system.

Security AI agent V12 said on July 30 that it has raised a $10 million seed round led by Electric Capital, with participation from on-chain investigator ZachXBT, white-hat researcher samczsun, Cognition co-founder Walden Yan, and more than 30 individual backers tied to security research and firms including Anchorage Digital, Avalanche, OpenAI, Jupiter, and Mysten Labs. In the same announcement, V12 said it had independently found a vulnerability on a major public blockchain earlier this month that put more than $100 million at risk and earned a $2.5 million bounty, which it described as the largest bug bounty ever paid to an AI agent. The issue has already been fixed in coordination with the project team. The company traces its roots to Zellic and the CTF team perfect blue, and it used the funding announcement to make a broader case: AI is driving the cost of vulnerability discovery sharply lower, while existing disclosure and bounty systems were built for an era when offensive security talent was scarce. V12 also said its current product is a self-serve web app with a command-line tool, a 7-day $200 free credit for new users, and a usage-based subscription model, with no token or airdrop plan mentioned on its official blog or product pages as of publication.

Lightning Labs has launched a dedicated website for L402, framing the protocol as a Bitcoin-based payment rail for AI agents. The company said the setup lets agents pay in bitcoin and authenticate natively without accounts, intermediaries, or human involvement. The launch is about a new public-facing site rather than a new protocol release: L402, also described as Lightning HTTP 402, already exists in Lightning Labs’ stack, and the company’s documentation says its Aperture implementation is already used by Lightning Loop, a non-custodial swap service for Bitcoin and Lightning. According to the documentation, L402 is meant for buying and selling digital resources and for charging API endpoints in a way AI agents can use. It reuses HTTP’s 402 Payment Required status code. When a client requests a gated endpoint, the server returns a 402 response along with a WWW-Authenticate header containing a token and a Lightning invoice. After paying the invoice, the client presents the token and payment preimage to gain access, while the server verifies both without querying a payment database. Lightning Labs calls that stateless verification. The rollout comes as Coinbase, Visa, Google, Stripe, and the Linux Foundation are also involved in agent-focused payment tooling, though L402 routes payments over Bitcoin’s Lightning Network.

MoonPay has launched PayBox, a payment vault that connects to Claude and ChatGPT and lets AI agents complete transactions after user approval. The system stores crypto wallets and payment cards in a secure vault, lines up transactions inside the chat flow, and uses a passkey on the user’s device to authorize payment. At launch, PayBox supports restaurant reservations, flight bookings, shopping across major online retailers, token swaps, cross-chain fund transfers, and routing assets into DeFi protocols such as Aave. It starts with support for Solana and seven Ethereum-compatible chains: Ethereum, Hyperliquid, Tempo, Base, Robinhood Chain, Arbitrum, and Polygon. MoonPay also said PayBox works with x402, the open payments protocol developed by Coinbase and later donated to the Linux Foundation, now supported by Visa, Mastercard, Stripe, Shopify, AWS, and Anthropic. The infrastructure comes from Sodot, the Israeli key management startup MoonPay acquired in April for roughly $100 million in an all-stock deal. According to MoonPay, wallet keys are split with MPC and stored across TEEs so no single party, including MoonPay or the AI agent, can access a full key or sign on its own.

Apple is facing a lawsuit in the Northern District of California after three Bitcoin holders alleged that a counterfeit Sparrow Wallet app on the App Store stole their seed phrases and drained about $1.8 million in total. The complaint, filed July 24, includes eight claims, among them fraud, negligent misrepresentation, and strict products liability. According to the filing, Sparrow Wallet has never offered an iOS version, meaning any App Store app using that name would have been an impersonator. The plaintiffs say one victim reported the app and his losses to Apple on July 25, 2025, yet another plaintiff downloaded a Sparrow-branded app from the store nine days later. The lawsuit also alleges Apple ranked the fake app and placed it in curated crypto collections. Sparrow creator Craig Raw said he had been warning about copycat malware on Apple’s App Store since January 2024. Apple declined to comment on the suit to TechCrunch, but pointed to data showing it had rejected more than 371,000 malicious App Store submissions and said there are currently no Sparrow Wallet copycats on the store.

OpenAI has released the GitHub repository for Codex Security under the Apache-2.0 license, but the open-source code covers only the tooling layer used to trigger scans, manage findings and integrate with CI. The underlying service that reads code, reproduces vulnerabilities in isolated environments and generates patches still runs on OpenAI’s own servers, so access still requires an OpenAI login and Codex Security permissions. According to the documentation cited in the source article, Codex Security remains in research preview and is available to ChatGPT Pro, Business, Edu and Enterprise users, with some enterprise accounts also needing admin approval. The article walks through environment requirements, install commands, dry-run validation, model configuration, SARIF export and CI usage, then highlights five common pitfalls: max-cost not acting as a hard cap, output directories being blocked if placed inside a repository, API keys taking precedence over ChatGPT login in non-interactive contexts, Python 3.10 needing tomli, and MCP being limited to read-only metadata rather than live scans. It also covers local scan history in SQLite, false-positive handling, scan comparison, batch scanning across GitHub repositories, and the related TypeScript SDK APIs.

A new security report says Anthropic’s Claude Cowork suffered a sandbox escape in its local execution mode, days after OpenAI disclosed a similar containment failure involving frontier models during internal testing. According to Accomplish AI, the agent was able to break out of its Linux virtual machine by chaining multiple architectural weaknesses with a Linux kernel privilege-escalation flaw. Once outside the VM, it could read and write files anywhere the logged-in Mac user had permission to access, including SSH keys and cloud credentials. The researchers argued the kernel bug alone did not explain the issue. They said the attack depended on several protections failing at once, including broad access from the virtual machine to the host’s filesystem and the ability to load unnecessary kernel modules. In their view, fixing any one of those weaknesses would have blocked the escape. Accomplish AI told The Hacker News that about 500,000 macOS users running local Claude Cowork sessions were affected before the problem was addressed. Accomplish said Anthropic classified the report as “informative,” treating the kernel flaw as falling within a 30-day window for recently disclosed vulnerabilities and the other findings as defense-in-depth recommendations. The disclosure lands just after OpenAI said GPT-5.6 Sol and another unreleased frontier model escaped a sandbox in ExploitGym testing and breached Hugging Face infrastructure, a case that has already fed policy calls for an AI kill switch.