209 posts tagged with "AI"

When Jensen Huang declared at CES 2026 that "the ChatGPT moment for physical AI is here," he was describing machines that understand, reason, and act in the real world. What he didn't say — but what a growing ecosystem of blockchain projects is betting on — is that those machines will also need to earn, spend, and own assets autonomously. Welcome to the era of DePAI: Decentralized Physical AI.

Over $3 million in agent-to-agent transactions were already happening on Ethereum — with no escrow, no delivery verification, and no recourse if something went wrong. On March 10, 2026, Virtuals Protocol and the Ethereum Foundation's dAI team submitted a proposal to fix that: ERC-8183, a new standard that turns raw on-chain payments between AI agents into verifiable, trustless commerce.

The timing is significant. The agentic AI market is projected to balloon from $7 billion in 2025 to $93 billion by 2032. Google launched its Universal Commerce Protocol in January 2026 with backing from Shopify, Walmart, Visa, and Mastercard. Coinbase's x402 protocol has processed over 35 million transactions on Solana alone. Yet none of these systems solve the fundamental trust problem that emerges when two autonomous programs try to do business with each other.

ERC-8183 does — and the way it does it may define how trillions of dollars in machine-to-machine commerce eventually settles.

An AI agent built by an OpenAI engineer accidentally sent $450,000 in tokens to a stranger on X who asked for $310 worth of SOL. No hack. No exploit. Just a session reset, a missing guardrail, and an irreversible blockchain transaction. The Lobstar Wilde incident in February 2026 was a wake-up call: if autonomous agents are going to handle real money, the industry needs a fundamentally different security model.

On March 13, 2026, MoonPay answered with one. Its CLI wallet now ships with native Ledger hardware signer support — making MoonPay Agents the first AI agent platform where every on-chain transaction must pass through a physical device before execution. Private keys never touch the agent runtime. The agent proposes; the human disposes.

When a human developer needs an API, they pull out a credit card, fill in a billing form, and start making calls. When an AI agent needs the same API, it hits a wall. No identity. No wallet. No way to pay. Sapiom's $15.75M seed round, led by Accel with backing from Anthropic, Coinbase Ventures, and Okta Ventures, is a bet that this wall is the single biggest bottleneck holding back the agentic economy — and that whoever tears it down will own the financial plumbing of a $3–5 trillion market.

What if the next frontier AI model wasn't trained in a billion-dollar data center owned by a single corporation — but by dozens of anonymous contributors scattered across the globe, coordinated by a blockchain, communicating over ordinary internet connections?

That's exactly what just happened. Templar's Covenant-72B, a 72.7-billion-parameter large language model pre-trained entirely on Bittensor's Subnet 3, has become the largest collaboratively trained AI model in crypto history — and one of the first to achieve competitive performance with centralized baselines while allowing fully permissionless participation. No whitelists. No corporate gatekeepers. Just GPUs, compressed gradients, and a token-incentive mechanism that kept everyone honest.

Anthropic co-founder Jack Clark called out the achievement in his influential Import AI newsletter, noting that decentralized training compute is growing at 20x per year — four times faster than centralized frontier training's 5x annual growth rate.

Here's why this matters far beyond the Bittensor ecosystem.

Aethir quietly crossed $127 million in annual revenue in 2025. Not in token emissions. Not in speculative incentive programs. In actual enterprise spending on GPU compute. That single data point may mark the moment decentralized compute stopped being a crypto experiment and started becoming a cloud business.

For years, the knock against Decentralized Physical Infrastructure Networks (DePIN) was simple: their economics ran on token printing, not customer invoices. Providers earned rewards denominated in volatile native tokens, demand was often synthetic, and the gap between "network activity" and "revenue" could be measured in orders of magnitude. But across 2025 and into early 2026, the leading GPU compute networks — Akash, io.net, Aethir, and Render — have been executing a pivot that the broader market hasn't fully priced in: the shift from token-subsidized supply to demand-driven cash flow.

When Andreessen Horowitz led a $30 million Series A into Lio on March 5, 2026, the enterprise software world took notice. But here's what caught many by surprise: Lio isn't another blockchain supply chain platform. It's an AI-powered agentic procurement system — and its success reveals where enterprise automation is actually heading in 2026.

The $180 Billion Manual Procurement Problem​

Enterprises spend over $180 billion annually on procurement talent, compared to roughly $10 billion on procurement software. That 18:1 ratio tells you everything you need to know about how broken corporate purchasing remains. Despite decades of ERP investments, procurement teams still manually chase quotes, negotiate terms, onboard vendors, and reconcile invoices across fragmented systems.

Lio's AI agents change the equation. Instead of incrementally improving existing workflows, the platform deploys specialized autonomous agents that work in parallel — researching vendors, negotiating terms, managing approvals, and tracking deliveries simultaneously. One global manufacturer automated 75% of its previously outsourced procurement operations within six months, achieving an 85% reduction in manual buyer work.

The funding round — which included participation from SV Angels, Harry Stebbings, and Y Combinator, bringing Lio's total capital to $33 million — reflects investor confidence that agentic AI, not blockchain, is the dominant automation paradigm for 2026 enterprise procurement.

AI Agents vs. Blockchain: The Enterprise Automation Divergence​

For years, blockchain evangelists pitched distributed ledger technology as the solution to supply chain opacity and procurement inefficiency. Smart contracts would automate payments. Immutable records would ensure compliance. Shared ledgers would eliminate reconciliation headaches.

Reality proved messier. While blockchain found traction in specific use cases — trade finance, multiparty settlement, provenance tracking for high-value goods — it struggled with the operational complexity of enterprise procurement. Consider the friction points:

Integration barriers: IBM Blockchain and Hyperledger Fabric require permissioned networks with pre-negotiated governance. Onboarding suppliers across heterogeneous ERP systems (SAP, Oracle, NetSuite) introduces months of technical overhead. Germany's Industrie 4.0 programs demonstrated blockchain-ERP integration is possible via APIs, but deployment remains confined to pilot-scale projects with willing participants.

Adoption chicken-and-egg: Blockchain's network effects require critical mass. A manufacturer can't tokenize purchase orders if suppliers aren't on-chain. The coordination problem stalls adoption — especially when existing EDI and API integrations already connect legacy systems.

Governance complexity: Who controls the blockchain? Who pays for nodes? How do you handle disputes when smart contracts execute incorrectly? These questions require legal frameworks that most enterprises haven't built.

Contrast that with Lio's AI agents. They operate within existing systems — ERPs, email inboxes, vendor portals, contract repositories — without requiring counterparties to adopt new infrastructure. Agents triage requests, analyze quotes, compare suppliers across the open web, and execute purchases end-to-end. The technology integrates with what you already have, rather than demanding rip-and-replace transformation.

The procurement software market is voting with its capital. In 2026, AI-driven platforms dominate enterprise automation investment, while blockchain supply chain projects remain concentrated in trade finance and compliance-heavy verticals like pharmaceuticals and luxury goods.

Why 94% of Procurement Executives Use AI Weekly (But Only 5% Reach Production Scale)​

By 2026, 94% of procurement executives use generative AI weekly, and 80% of Chief Procurement Officers prioritize AI investments at the strategy level. Yet here's the paradox: over 80% of enterprise firms pilot generative AI, but only 5% of AI pilots reach mature production-stage adoption.

What explains the gap?

Deployment maturity lags hype. Most 2024-2025 AI procurement pilots focused on narrow use cases: contract summarization, spend classification, basic chatbots. These tools delivered marginal improvements but didn't fundamentally restructure workflows. Executives got incremental gains, not transformation.

Agentic AI changes the equation. Unlike template-based automation, agentic AI handles end-to-end tasks and exceptions autonomously. Lio's agents don't just summarize contracts — they source vendors, negotiate terms, and execute purchases. The shift from "AI as assistant" to "AI as workforce" represents the maturity leap enterprises need to cross the 5% production threshold.

Enterprise procurement remains stubbornly manual. Even advanced ERP systems require human coordination across purchasing, legal, finance, and operations. Lio's multi-agent architecture parallelizes these workflows. One agent researches suppliers while another evaluates compliance while a third negotiates pricing. The compound efficiency gains justify serious capital investment.

The $30 million Lio raise signals that investors believe 2026 is the inflection year when agentic AI moves from pilot curiosity to production infrastructure.

Blockchain's Niche: Where DLT Still Wins in Procurement​

Blockchain hasn't disappeared from enterprise procurement — it's finding its niche. Market projections estimate supply chain blockchain applications could surpass $15 billion in value by 2026, growing from $1.17 billion in 2024 to a projected $33.25 billion by 2033 at a 39.7% CAGR.

Where is blockchain actually delivering ROI?

Trade finance and multiparty settlement. When multiple parties need shared, immutable transaction records — especially across jurisdictions with limited trust — blockchain provides value. Banks, customs authorities, shippers, and importers use platforms like TradeLens and Marco Polo to reduce reconciliation costs and fraud.

Provenance and compliance. Luxury goods manufacturers use blockchain to prove authenticity. Pharmaceutical companies track temperature-sensitive shipments. Organic food supply chains verify certifications. These use cases share a common pattern: high-value goods where verifiable provenance justifies the integration overhead.

Smart contract automation in regulated contexts. When contractual terms are standardized and regulatory frameworks demand auditability, blockchain-based smart contracts offer advantages. Payment-on-delivery triggers, escrow arrangements, and multi-signature approvals reduce manual intervention.

Blockchain excels when trust is scarce, verification is valuable, and counterparties are willing to adopt shared infrastructure. AI agents excel when speed matters, integration complexity is high, and workflows span heterogeneous systems.

The Web3 Angle: Why Blockchain Infrastructure Matters Even If Procurement Goes AI-First​

For Web3 infrastructure providers, Lio's success might seem like a validation of AI over blockchain. But the story is more nuanced.

First, blockchain-ERP integration is advancing. Wholechain and other traceability platforms are connecting permissioned DLTs to SAP and Oracle systems, proving that enterprise blockchain isn't dead — it's maturing. The integration of blockchain with cloud platforms and alignment with GDPR, HIPAA, and sector-specific compliance rules are cutting reconciliation costs and reducing fraud and audit risk.

Second, the AI agent economy will need blockchain rails. As Lio-style AI agents proliferate, they'll increasingly transact with each other — purchasing compute resources, licensing data, settling micropayments for API calls. Web3's programmable payment infrastructure (stablecoins, smart contracts, decentralized identity) could become the financial plumbing for autonomous agent-to-agent commerce.

Third, hybrid architectures are emerging. Deloitte's research on blockchain-driven supply chain innovation highlights how enterprises are combining AI analytics with blockchain transparency. AI agents optimize purchasing decisions; blockchain provides immutable audit trails. The technologies complement rather than compete.

What Lio's $30M Means for Enterprise Automation in 2026​

Three takeaways emerge from Lio's funding round:

1. Agentic AI is entering production. The shift from pilots to deployed workflows is happening now. Lio's claim that it manages "billions in spend" for 100+ clients — including Fortune 500 companies — demonstrates real traction beyond proof-of-concept. Expect more AI agent platforms to raise serious capital in 2026.

2. Integration trumps ideology. Enterprises don't care whether the technology is blockchain, AI, or traditional automation — they care about ROI, deployment speed, and compatibility with existing systems. AI agents win procurement because they integrate with what's already there. Blockchain wins trade finance because counterparties accept shared ledgers. Technology choice follows business logic, not hype.

3. The $180 billion manual procurement market is up for grabs. If AI can automate 75-85% of procurement work, the talent spend collapses and software spend explodes. Lio's Series A is the opening salvo in a land grab for enterprise purchasing automation. Competitors will emerge, incumbents will respond, and M&A will consolidate the space.

For Web3 builders, the lesson isn't "blockchain lost." It's that enterprise adoption follows value, not narrative. Blockchain infrastructure that delivers ROI in specific contexts — trade finance, compliance, provenance — will thrive. But expecting every enterprise workflow to run on-chain was always a fantasy.

The 2026 Enterprise Automation Landscape​

As we move deeper into 2026, the enterprise automation landscape is bifurcating:

AI-first workflows: Procurement, customer service, financial analysis, HR onboarding — anywhere speed and integration matter more than trust guarantees.

Blockchain-first workflows: Trade settlement, provenance tracking, multiparty compliance — anywhere verifiable shared state matters more than deployment speed.

Hybrid systems: Supply chain visibility (AI analytics + blockchain transparency), tokenized securities (AI risk models + on-chain settlement), cross-border payments (AI fraud detection + stablecoin rails).

Lio's $30 million raise confirms that 2026 belongs to AI agents in procurement. But the story doesn't end there. As agent economies scale, they'll need Web3 infrastructure for identity, payments, and programmable coordination.

The question for blockchain builders: are you building for enterprises that want incremental automation? Or for the autonomous agent economy that doesn't exist yet but is coming fast?

---

Enterprise automation is evolving rapidly, and the infrastructure layer is critical. Whether you're building AI-driven workflows or blockchain-based settlement systems, reliable API access is non-negotiable. Explore BlockEden.xyz's enterprise-grade infrastructure services for blockchain and Web3 integrations built to scale.

Sources​

• Lio raises $30M from Andreessen Horowitz and others to automate enterprise procurement | TechCrunch
• Lio Raises $30M Series A to Bring Agentic AI to Enterprise Procurement | PR Newswire
• Investing in Lio | Andreessen Horowitz
• State of AI in Procurement in 2026 | Art of Procurement
• Using Blockchain to Drive Supply Chain Transparency and Innovation | Deloitte US
• Blockchain for Supply Chain in 2026 | ScienceSoft
• Enhancing Supply Chain Transparency with Blockchain-Enabled ERP Systems | Medium
• Understanding Blockchain in Supply Chain & Procurement | GEP

GitHub's fastest-rising repository in history just exposed over 135,000 vulnerable AI agents across 82 countries—and crypto users are the primary targets. Welcome to the OpenClaw security crisis, where Chinese tech giants racing to deploy AI gateways collided with a massive supply chain attack that's rewriting the rules for blockchain security.

The Viral Phenomenon That Became a Security Nightmare​

In late January 2026, OpenClaw achieved something unprecedented: it gained over 20,000 GitHub stars in a single day, becoming the platform's fastest-growing open-source project ever. By March 2026, the AI assistant had amassed over 250,000 stars, with tech enthusiasts worldwide rushing to install what seemed like the future of personal AI.

Unlike cloud-based AI assistants, OpenClaw runs entirely on your computer with full access to your files, email, and applications. You can message it through WhatsApp, Telegram, or Discord, and it works 24/7—executing shell commands, browsing the web, sending emails, managing calendars, and taking actions across your digital life—all triggered by a casual message from your phone.

The pitch was irresistible: your own personal AI agent, running locally, always available, infinitely capable. The reality turned out to be far more dangerous.

135,000 Exposed Instances: The Scale of the Security Disaster​

By February 2026, security researchers discovered a chilling fact: more than 135,000 OpenClaw instances were exposed on the public internet across 82 countries, with over 50,000 vulnerable to remote code execution. The cause? A fundamental security flaw in OpenClaw's default configuration.

OpenClaw binds by default to 0.0.0.0:18789, meaning it listens on all network interfaces including the public internet, rather than 127.0.0.1 (localhost only) as security best practices demand. For context, this is equivalent to leaving your front door wide open with a sign saying "enter freely"—except the door leads to your entire digital life.

The "ClawJacked" vulnerability made the situation even worse. Attackers could hijack your AI assistant simply by getting you to visit a malicious website. Once compromised, the attacker gains the same level of access as the AI agent itself: your files, credentials, browser data, and yes—your crypto wallets.

Security firms scrambled to understand the scope. Kaspersky, Bitsight, and Oasis Security all issued urgent warnings. The consensus was clear: OpenClaw represented a "security nightmare" involving critical remote code execution vulnerabilities, architectural weaknesses, and—most alarmingly—a large-scale supply chain poisoning campaign in its plugin marketplace.

ClawHavoc: The Supply Chain Attack Targeting Crypto Users​

While researchers focused on OpenClaw's core vulnerabilities, a more insidious threat was unfolding in ClawHub—the marketplace designed to make it easy for users to find and install third-party "skills" (plugins) for their AI agents.

In February 2026, security researchers codenamed ClawHavoc discovered that out of 2,857 skills audited on ClawHub, 341 were malicious. By mid-February, as the marketplace grew to over 10,700 skills, the number of malicious skills had more than doubled to 824—and by some reports, reached as high as 1,184 malicious skills.

The attack mechanism was devastatingly clever:

1. Fake prerequisites: 335 skills used fake installation requirements to trick users into downloading the Atomic macOS Stealer (AMOS) malware
2. Platform-specific payloads: On Windows, users downloaded "openclaw-agent.zip" from compromised GitHub repositories; on macOS, installation scripts hosted at glot.io were copied directly into Terminal
3. Sophisticated social engineering: Documentation convinced users to execute malicious commands under the guise of legitimate setup steps
4. Unified infrastructure: All malicious skills shared the same command-and-control infrastructure, indicating a coordinated campaign

The primary targets? Crypto users.

The malware was designed to steal:

• Exchange API keys
• Wallet private keys
• SSH credentials
• Browser passwords
• Crypto-specific data from Solana wallets and wallet trackers

Out of the malicious skills, 111 were explicitly crypto-focused tools, including Solana wallet integrations and cryptocurrency trackers. The attackers understood that crypto users—accustomed to installing browser extensions and wallet tools—would be the most lucrative targets for an AI agent supply chain attack.

The Chinese Tech Giant Deployment Race​

While security researchers issued warnings, Chinese tech giants saw opportunity. In early March 2026, Tencent, Alibaba, ByteDance, JD.com, and Baidu all launched competing free OpenClaw installation campaigns, compressing a competitive scramble that typically takes months into just days.

The strategy was clear: use free deployments as customer acquisition, locking in users before commercial AI projects scale up. Each giant raced to become the "first infrastructure contact for the next generation of AI developers":

• Tencent launched QClaw, integrating OpenClaw with WeChat so users could remotely control their laptops by sending commands via their phones
• Alibaba Cloud rolled out support for OpenClaw across its platforms, connecting to its Qwen AI model series
• ByteDance's Volcano Engine unveiled ArkClaw, an "out-of-the-box" version of OpenClaw

The irony was stark: as security researchers warned of 135,000 exposed instances and massive supply chain attacks, China's largest tech companies were actively promoting mass installation to millions of users. The collision between technological enthusiasm and security reality had never been more visible.

Web3's AI Agent Problem: When MCP Meets Crypto Wallets​

The OpenClaw crisis exposed a deeper issue that Web3 builders can no longer ignore: AI agents are increasingly managing on-chain assets, and the security models are dangerously immature.

The Model Context Protocol (MCP)—the emerging standard for connecting AI agents to external systems—is becoming the gateway through which AI interacts with blockchains. MCP servers function as unified API gateways to the full Web3 stack, enabling AI agents to read blockchain data, prepare transactions, and execute on-chain actions.

Currently, most cryptocurrency MCP servers require configuration with a private key, creating a single point of failure. If an AI agent is compromised—as tens of thousands of OpenClaw instances were—the attacker gains direct access to funds.

Two competing security models are emerging:

1. Delegated Signing (User-Controlled)​

AI agents prepare transactions, but the user retains exclusive control over signing. The private key never leaves the user's device. This is the most secure approach but limits agent autonomy.

2. Agent-Controlled Allowances​

Agents have their own keys and receive an allowance to spend on behalf of users. Private keys are managed securely by the agent host, and spending is capped. This enables autonomous operation but requires trust in the host's security.

Neither model is widely adopted yet. Most crypto MCP implementations still use the dangerous "give the agent your private key" approach—exactly the scenario ClawHavoc attackers were counting on.

By 2026 estimates, 60% of crypto wallets will use agentic AI to manage portfolios, track transactions, and improve security. The industry is implementing Multi-Party Computation (MPC), account abstraction, biometric authentication, and encrypted local storage to secure these interactions. Standards like ERC-8004 (co-led by the Ethereum Foundation, MetaMask, and Google) are attempting to create verifiable identity and credit history for AI agents on-chain.

But OpenClaw proved these safeguards aren't in place yet—and attackers are already exploiting the gap.

NVIDIA's Enterprise Answer: NemoClaw at GTC 2026​

As the OpenClaw security crisis unfolded, NVIDIA saw an opening. At GTC 2026 in mid-March, the company announced NemoClaw, an open-source AI agent platform specifically designed for enterprise automation with security and privacy built in from the ground up.

Unlike OpenClaw's consumer-first, install-anywhere approach, NemoClaw targets businesses with:

• Built-in security and privacy tools addressing the vulnerabilities that plagued OpenClaw
• Enterprise authentication and access controls preventing the "open to the internet" default configuration disaster
• Multi-platform support that runs beyond just NVIDIA chips, leveraging the company's NeMo, Nemotron, and Cosmos AI frameworks
• Partnership ecosystem including talks with Salesforce, Google, Cisco, Adobe, and CrowdStrike

The timing couldn't be more strategic. As OpenClaw's "Lobster Fever" exposed the dangers of consumer-focused AI agents, NVIDIA positioned NemoClaw as the secure, enterprise-grade alternative—potentially challenging OpenAI in the business AI agent market.

For Web3 companies building AI-integrated infrastructure, NemoClaw represents a potential solution to the security problems OpenClaw exposed: professionally managed, audited, and secured AI agent deployments that can safely interact with high-value blockchain assets.

The Wake-Up Call Web3 Needed​

The OpenClaw crisis isn't just an AI security story—it's a blockchain infrastructure story.

Consider the implications:

• 135,000+ exposed AI agents with potential access to crypto wallets
• 1,184 malicious plugins specifically targeting cryptocurrency users
• Five Chinese tech giants pushing millions of installations without adequate security review
• 60% of crypto wallets projected to use AI agents by year-end
• No widely adopted security standards for AI-blockchain interactions

This is Web3's "supply chain security moment"—comparable to the 2020 SolarWinds attack in TradFi or the 2016 DAO hack in crypto. It exposes a fundamental truth: as blockchain infrastructure becomes more powerful and automated, the attack surface expands exponentially.

The industry's response will define whether AI agents become a secure gateway to Web3 functionality or the largest vulnerability the space has ever seen. The choice between delegated signing models, agent allowances, MPC solutions, and account abstraction isn't just technical—it's existential.

What Web3 Builders Should Do Now​

If you're building in Web3 and integrating AI agents—or planning to—here's the checklist:

1. Audit your MCP server security: If you're requiring private keys for AI agent access, you're creating ClawHavoc-style attack vectors
2. Implement delegated signing: Users should always retain exclusive control over transaction signing, even when AI prepares transactions
3. Use allowance-based models for autonomous agents: If agents need to act independently, give them dedicated keys with strict spending limits
4. Never install AI agents with default network configurations: Always bind to localhost (127.0.0.1) unless you have enterprise-grade authentication
5. Treat AI agent marketplaces like app stores: Require code signing, security audits, and reputation systems before trusting third-party skills
6. Educate users about AI agent risks: Most crypto users don't understand that an AI agent is functionally equivalent to giving someone root access to their computer

The OpenClaw crisis taught us that security-by-default matters more than features. The race to deploy AI agents can't outpace the race to secure them.

Building blockchain infrastructure that connects to AI agents? BlockEden.xyz provides enterprise-grade API infrastructure for over 40 blockchains with security-first architecture designed for high-stakes integrations. Explore our services to build on foundations designed to last.

---

Sources:

• Meet OpenClaw: The AI assistant that broke every record – and started a security panic
• OpenClaw: GitHub's Fastest-Ever Rising Star Becomes 2026's First Major AI Security Disaster
• New OpenClaw AI agent found unsafe for use | Kaspersky official blog
• OpenClaw Security: Risks of Exposed AI Agents Explained | Bitsight
• ClawJacked: OpenClaw Vulnerability Enables Full Agent Takeover
• Researchers Find 341 Malicious ClawHub Skills Stealing Data from OpenClaw Users
• Malicious OpenClaw Skills Used to Distribute Atomic MacOS Stealer | Trend Micro
• ClawHavoc Poisons OpenClaw's ClawHub With 1,184 Malicious Skills
• Using MCP with Web3: How to secure blockchain-interacting agents | Google Cloud Blog
• How to Build AI-Powered Web3 Crypto Wallets in 2026
• Model Context Protocol (MCP): Bridging AI and Blockchain for Smarter Crypto Projects
• Chinese Tech Giants Race to Adopt OpenClaw AI Gateway
• China's AI Red Packet War: ByteDance, Tencent, Alibaba and Baidu Battle to Become the Default Gateway
• Nvidia targets enterprise AI agents with new open-source NemoClaw platform
• Nvidia Prepares NemoClaw to Power Enterprise AI Agents
• NVIDIA to Launch Open-Source AI Agent "NemoClaw" at GTC 2026: What We Know So Far

Circle, Stripe, and Coinbase are spending billions building payment rails for AI agents that barely transact. Is this the next internet moment — or the next metaverse mirage?