Base Halts for Two Hours: A Single Invalid Block Reveals the Centralized Reality of L2s

Foresight NewsPublished on 2026-06-26Last updated on 2026-06-26

Abstract

Base, an Ethereum Layer-2 Rollup, experienced a two-hour network outage starting around 00:03 UTC on June 26. The halt was caused by a consensus issue that led to an invalid block being sequenced, which prevented the generation of new blocks after block 47806542. The team identified the problem, restored block sequencing by 01:51 UTC, and confirmed full recovery of ecosystem infrastructure synchronization shortly after. This incident highlights the operational reality for many L2s: while they leverage Ethereum for security and data availability, their day-to-day usability heavily depends on their sequencer and internal systems. Base employs a high-availability sequencer system with one active leader, but this setup did not prevent the outage when a consensus-level problem arose. This follows a previous 33-minute outage in August 2025 related to a faulty sequencer handover process. The downtime occurred near the scheduled activation window for the "Beryl" network upgrade, which has since been postponed. Beryl introduces the native B20 token standard, among other improvements. The incident has sparked renewed discussion about Base potentially launching its own network token in the future, shifting the conversation from mere speculation to questions about how a token might relate to sequencer decentralization, governance, and accountability in such failure scenarios.

By: KarenZ, Foresight News

In the early hours of June 26th Beijing time, Base gave the market a quiet yet profound lesson on infrastructure.

The timeline of this outage is clear:

At 0:03, Base reported abnormal block production status on the mainnet, with the team investigating.
At 0:52, Base identified the issue as stemming from a problematic block that would disrupt the construction of subsequent blocks.
At 1:21, Base located a consensus issue that led to an invalid block being sequenced. This prevented the generation of new blocks after block 47806542. The internal sequencer and nodes had been preliminarily restored.
At 1:51, the sequencing of new blocks resumed, and internal node synchronization was normal.
At 1:58, Base confirmed that healthy block construction had resumed, and ecosystem infrastructure was able to resume synchronization.
At 3:22, Base further stated that the sequencer and related systems remained stable, blocks were generating normally, and the team had identified the root cause of the outage. They were validating a fix and would later publish a full post-mortem report.

Base's status page indicated this incident affected deposits, withdrawals, block production, and client software on the Base mainnet.

The Single Active Sequencer Makes the Halt More Glaring

Base is a Rollup built on Ethereum. It executes a large volume of transactions on the L2 and then submits necessary data and state-related information to Ethereum.

What users perceive daily is not the architecture diagram, but whether transactions can get into a block, nodes can sync, and wallets, trading, and cross-chain services function normally.

Before Flashblocks went live, Base used a highly available sequencer system: among 5 sequencer instances, one served as the leader, responsible for building blocks and propagating them via P2P, while the other 4 acted as followers, syncing chain state. If the current leader stopped producing blocks, the system would initiate a leadership switch.

This design shows Base is not without redundancy. The issue is that redundancy primarily addresses failover and availability, which is not equivalent to having multiple independent sequencers simultaneously participating in block production. Daily block production is still undertaken by the current leader. Once problems arise with consensus, sequencing, or node synchronization pathways, the first thing users experience is the halting progress of new blocks.

After Flashblocks went live, Base's block building gained an additional layer of sub-second pre-confirmation mechanism. Base documentation states that Flashblocks are always enabled on Base; all blocks are built by the Flashblocks builder. Applications can choose whether to use pre-confirmation data or continue waiting for the standard 2-second block confirmation via standard RPC. In other words, Flashblocks is already part of the infrastructure for Base's current block building, but it's an optional feature for applications seeking the pre-confirmation experience.

Base's security status page provided a more specific explanation: a consensus issue led to an invalid block being sequenced, which in turn prevented the generation of new blocks after block 47806542. The exact root cause still awaits the official post-mortem.

The Last Base Outage Was Due to Inability to Successfully Configure a New Sequencer

This is not the first time Base's block production has been interrupted due to sequencer-related issues. On August 5, 2025, the Base mainnet experienced a 33-minute network outage. The official post-mortem stated that the then-active sequencer began experiencing delays due to on-chain activity; Conductor, responsible for managing the high-availability (HA) cluster, automatically switched leadership to a new sequencer. However, the new sequencer was still in the configuration process at the time and could not produce blocks. Furthermore, because Conductor was not yet fully enabled on that sequencer, the system could not initiate another switch. The team then manually paused HA, transferred leadership to a healthy sequencer, and later fully restored operations.

Comparing the two incidents requires caution: the August 2025 issue pointed to flaws in the sequencer high-availability failover process, while today's outage was caused by a consensus problem that led to an invalid block being sequenced, preventing new blocks after 47806542.

Together, they point to a reality: L2s can rely on Ethereum for core issues like data availability, settlement, security, and finality, but their daily availability hinges heavily on the sequencer and related operational systems. As long as new blocks cannot be generated, users see transactions stuck on the road.

This Outage Occurred Close to the B20 Launch Window

The timing of the incident is delicate. Base was in the vicinity of the Beryl upgrade window at the time.

The original mainnet activation time for Beryl was 2:00 AM Beijing time on June 26th, now postponed to 2:00 AM on June 27th.

Beryl's core features include three items: introducing the B20 native token standard, reducing the single-proof withdrawal finalization period from 7 days to 5 days, and offering up to a 50% reduction in disk usage and approximately 33% throughput improvement via Reth V2.

B20 differs in its underlying implementation. Most ERC-20 tokens are deployed as EVM smart contracts, whereas B20 is implemented via Rust precompile and created through the singleton B20Factory. It also includes built-in capabilities like role-based permissions, supply caps, mint/burn, pausing, transfer policies, memo, and ERC-2612 permit support. In simpler terms, Base has packaged many foundational token functions that issuers repeatedly build, audit, and maintain into a chain-level toolbox.

B20 most easily sparks market speculation, with some community users even associating it with Base launching its own token. However, B20 is about "how others can issue assets on Base in a more standardized way," while the discussion about Base launching a token concerns "whether Base will introduce its own network token in the future."

The former is already written into the Beryl upgrade; the latter remains a market concern without official announcement.

This outage will make the discussion about Base's potential token more realistic. The market used to ask: will Base launch a token, when, and how will the airdrop be distributed? After this incident, a more pertinent question is: if a network token is indeed introduced in the future, what responsibilities should it correspond to? Should it govern sequencer decentralization, governance constraints, a security budget, or the allocation of rights and responsibilities in incident response?

NVIDIA's New Open-Source MoE: One Line of Import, Fine-Tuning Accelerated by 3.7x

NVIDIA has open-sourced NeMo AutoModel, a tool designed to significantly accelerate the fine-tuning of Mixture-of-Experts (MoE) large language models. By adding just one import line to existing code based on Hugging Face Transformers v5, users can achieve a 3.4x to 3.7x increase in training throughput and reduce GPU memory usage by 29% to 32% without altering their API. The key innovations include Expert Parallelism (EP) to distribute expert weights across GPUs, lowering memory pressure; DeepEP to fuse computation and communication; and TransformerEngine kernels for accelerated core operations. Benchmarks on models like Qwen3-30B-A3B show training throughput per GPU jumping from 3075 to 11340 tokens per second. The solution also enables the fine-tuning of very large models, such as the 550B parameter Nemotron 3 Ultra, which would exceed memory limits with the standard Transformers v5. Code and benchmarks are available on GitHub.

marsbit27m ago

NVIDIA's New Open-Source MoE: One Line of Import, Fine-Tuning Accelerated by 3.7x

marsbit27m ago

How to Detect AI-Generated Videos? A Review of Dynamic, Traceable, and Explainable Detection Systems

**How to Detect AI-Generated Videos: A Survey on Dynamic, Traceable, and Explainable Detection Systems** With rapid advances in AI video generation (e.g., Sora, Veo), creating highly realistic, multi-minute videos is now possible, widening the gap with detection research. Current AI video detection, often limited to unreliable binary classifications, is insufficient. This survey, accepted at ACL 2026, reframes the goal as **"factual fidelity verification"**—checking if a video's content (who, when, where, what) aligns with the real world perceptually and cognitively. It categorizes AI-generated videos into three paradigms: **Local Manipulation Videos (LMV**, e.g., face swaps), **Audio-Visual Editing (AVE**, e.g., lip-syncing), and **Generative Video Synthesis (GVS**, fully synthetic videos like Sora's). Detection challenges evolve from visual artifacts in LMV to multi-modal inconsistencies in AVE and higher-level world knowledge violations in GVS. The core proposal is a **Vision-Language Dual-View framework** with four hierarchical layers: 1. **Layer 1 (Intrinsic Visual Cues):** Analyzes low-level signal statistics, noise patterns, and physiological signals. 2. **Layer 2 (Spatiotemporal Consistency):** Checks for temporal coherence in object motion and scene dynamics. 3. **Layer 3 (Cross-Modal Consistency):** Verifies alignment between video, audio, and text within the video. 4. **Layer 4 (Language-Guided World-Level Reasoning):** Uses external knowledge, facts, and physical laws to judge semantic plausibility and factual correctness. The survey traces a shift in detection focus from lower layers (1 & 2) toward higher, language-involved layers (3 & 4). It also reviews evolving evaluation metrics and datasets tailored for each video paradigm. The conclusion advocates for a **dynamic, evidence-first detection system** that moves beyond simple classification. Future trustworthy detection requires combining visual evidence (from CV) with semantic reasoning and explanation (from NLP & multimodal AI), ultimately creating traceable and explainable judgments about a video's adherence to real-world constraints.

marsbit29m ago

How to Detect AI-Generated Videos? A Review of Dynamic, Traceable, and Explainable Detection Systems

marsbit29m ago

It Turns Out the First Real-World Application of AI x Crypto is in Security Auditing

The article explores the surprising trend where AI's first major impact on crypto has been in security auditing, not in areas like trading or analytics. It details how AI-powered tools are dramatically lowering the barrier to finding smart contract vulnerabilities, enabling attackers to scan thousands of contracts and execute exploits within minutes. This has rendered traditional, manually-produced audit reports with their month-long validity periods increasingly obsolete, creating a critical "structural crack" in the old security model. Cases like Drift Protocol and KelpDAO show that even extensively audited protocols can be hacked through social engineering, operational flaws, or infrastructure misconfigurations beyond pure code review. Attackers are also using AI to find and exploit vulnerabilities in years-old, deployed contracts. Notably, OpenZeppelin's co-founder has expressed a grim view that "all DeFi is insecure" due to AI's asymmetric advantage. In response, the audit industry is undergoing a fundamental shift. While there's a short-term spike in defensive re-audits, the long-term business model is changing. Firms are developing AI-assisted systems and moving from one-time report deliveries towards embedded, continuous services like real-time monitoring and formal verification. Examples include AI tools uncovering critical, previously missed vulnerabilities in heavily audited protocols like Curve Finance and Zcash. The conclusion is that security must become a continuous investment, not a one-time checkbox, and audit firms must rapidly evolve their tools and service models to survive.

marsbit36m ago

It Turns Out the First Real-World Application of AI x Crypto is in Security Auditing

marsbit36m ago

Never expected that the first tangible application of AI x Crypto is in security auditing

Unexpectedly, the initial major application of AI in the Crypto sphere has turned out to be security auditing. In 2026, DeFi has faced significant security challenges, with 121 hacking incidents resulting in approximately $942 million in losses. While AI was expected to first impact areas like quantitative trading, its initial breakthrough has instead transformed security auditing by drastically lowering the cost and skill barrier for finding smart contract vulnerabilities. The traditional audit model is facing obsolescence. Advanced AI models, such as Claude Mythos, enable attackers to scan thousands of contracts and identify vulnerability patterns at scale, compressing the time from discovery to execution to mere minutes. This renders the month-long validity of traditional audit reports ineffective. Notably, attacks now frequently target well-audited, established protocols by exploiting business logic flaws, operational security weaknesses, and even years-old historical contracts, demonstrating that old audit reports offer zero protection. This pressure is forcing a fundamental shift in the industry. In the short term, a wave of defensive re-auditing is occurring, driven by projects seeking to meet new AI-era security standards and regulatory requirements. In the long run, audit firms' business models are diverging. The one-time report delivery model is declining in value, as evidenced by platforms like Code4rena shutting down. Leading firms are now pivoting towards AI-powered defense, integrating continuous monitoring, real-time on-chain risk detection, and embedding security directly into the development phase, as seen with tools like OpenZeppelin's Skills system. Ultimately, the era of "audit once, secure forever" is over. Security must become a continuous, embedded infrastructure investment for projects. For audit companies, survival depends on proactively transforming from traditional service providers into platforms offering AI-native, ongoing security solutions.

链捕手43m ago

Never expected that the first tangible application of AI x Crypto is in security auditing

链捕手43m ago

$3M Exploit Hits Polymarket: Users to Receive Full Refunds After Third-Party Breach

Polymarket will fully reimburse users affected by a front-end exploit that stole approximately $3 million in crypto assets. The company stated the incident was not a flaw in its core protocol but a supply chain attack. A compromised third-party vendor injected a malicious script into the platform's front-end for a limited number of users, allowing the attacker to drain funds from about 15 wallets while they interacted with the site. The stolen assets, primarily Polymarket's pUSD stablecoin, were bridged to Ethereum and converted to roughly 1,893 ETH. This marks the second security incident for Polymarket in under two months, following a separate $700,000 loss. The event highlights growing risks associated with third-party dependencies in the crypto industry.

TheNewsCrypto44m ago

$3M Exploit Hits Polymarket: Users to Receive Full Refunds After Third-Party Breach

TheNewsCrypto44m ago

Trading

Spot