Today in The Chain Reactor: The exodus from LayerZero accelerates as another $700M migrates, and we're seeing the first signs of AI's absolute monopoly on venture capital cooling off. Also in the mix: the developer toolchain shifts toward 'loop engineering', a $4.7M bridge exploit sits unnoticed for a week, and mainstream tech giants throw their weight behind the CLARITY Act.
Sophia Space, a Pasadena startup founded by JPL veteran Dr. Leon Alkalai, has closed an $11.5 million seed round to develop orbital data centers. The company is leveraging technology from JPL and Caltech that uses passive cooling in the vacuum of space, aiming to create foundational infrastructure for the burgeoning space economy.
Why it matters
This is a classic LA deep tech play, combining aerospace and data infrastructure. Building data centers in orbit could solve the immense power and cooling challenges facing terrestrial AI compute, while also enabling real-time data processing for satellites. It's a high-risk, high-reward bet on the future of compute and a perfect example of the unique convergence of talent and technology in the Southern California ecosystem.
Google researchers have published a paper proposing a new architecture that could challenge the dominance of transformers in LLMs. Titled 'Memory Caching: RNNs with Growing Memory,' the approach gives Recurrent Neural Networks (RNNs) an expanding memory capacity, similar to transformers, but with greater efficiency. It achieves this by storing compressed memory checkpoints instead of every token, creating a hybrid system.
Why it matters
This research is significant because it points to a potential architectural path beyond the compute-heavy, attention-based models that currently define the frontier. For a startup engineer, this is a signpost for where model architecture might be heading next. A system that offers the recall of transformers with the efficiency of RNNs could dramatically lower inference costs and enable more powerful models to run on-device, changing the economic and technical calculus for building AI products.
Perplexity has launched 'Brain,' a new memory system for its AI platform that enables agents to learn and improve over time. The system records and analyzes past tasks, successes, and failures, using that data to continuously update an internal LLM-powered knowledge base. This allows the AI to develop expertise and institutional knowledge rather than starting every session from a blank slate.
Why it matters
This moves AI agents beyond static, stateless responders toward systems with continuity and memory. It's a key step towards true adaptive intelligence. For anyone building AI products, this highlights a new competitive axis: the quality of an agent's memory and its ability to learn from experience will become as important as the underlying model's capabilities. It's a shift from 'how smart is the model?' to 'how fast can the system learn?'.
A new consensus methodology called 'loop engineering' has emerged as the default production pattern for agentic coding. This approach moves beyond simple one-shot prompts, instead building autonomous systems where AI agents plan, execute, test, and retry complex tasks with minimal human intervention. Boris Cherny, co-founder of Anthropic, is now a vocal proponent, stating the days of manual prompting are over in favor of AI agents that generate and refine their own prompts within these loops.
Why it matters
This marks a fundamental evolution in how engineers use AI, shifting the focus from prompt crafting to designing, observing, and managing autonomous workflows. For a startup engineer, this is the new frontier of developer tooling. Mastering the construction of these agentic loops—where the AI manages the entire development cycle for a given task—promises significant gains in velocity and efficiency, allowing small teams to tackle larger, more complex software projects.
Coinbase's Layer 2, Base, will activate its Beryl hard fork on mainnet on June 25. The upgrade introduces the B20 native token standard, designed to allow compliant asset issuance (like stablecoins and RWAs) at the protocol level. B20 tokens are implemented as Rust precompiles within the node software itself. The upgrade also integrates the Reth V2 execution client, promising up to 50% less disk usage and 33% more throughput, and shortens the standard withdrawal period to Ethereum from seven to five days.
Why it matters
This is a strategic move by Base to become the go-to chain for institutional and regulated token issuers. By embedding compliance-friendly features directly into the protocol with the B20 standard, Base is lowering the friction for big players to bring assets on-chain. For developers, this creates a new, more robust primitive for building financial applications and signals a maturing of L2 architecture beyond just scaling transactions to providing specialized financial infrastructure.
DeFi protocol Hyperliquid has completed its transition from a testnet to its own sovereign Layer 1 mainnet, called Hyperliquid EVM, which went live this week. The chain uses a custom 'HyperBFT' consensus mechanism designed for high throughput and sub-second finality. By adding EVM compatibility, developers can now deploy Solidity smart contracts, turning the previously specialized perpetuals exchange into a general-purpose L1.
Why it matters
This is a prime example of the 'app-chain' thesis playing out: a successful application building its own specialized blockchain to escape the constraints of a general-purpose environment. Adding EVM compatibility is the key second step, allowing them to leverage the massive Ethereum developer community. For engineers, this trend creates a new set of platforms that blend high performance with familiar tooling, offering compelling alternatives for deploying demanding DeFi applications.
An attacker drained approximately $4.67 million from an Axelar bridge connected to Secret Network by exploiting a vulnerability in a modified IBC contract. The attack, which happened on Saturday, involved minting unbacked wrapped tokens on Secret and redeeming them for real assets on Axelar. The exploit went completely undetected for seven days until a failed transfer attempt by another user exposed the now-empty escrow account. Axelar has since shut down the affected bridge routes.
Why it matters
This is a brutal reminder of the fragility of cross-chain bridges and the glaring gaps in monitoring within DeFi. The fact that nearly $5 million could be siphoned off with no one noticing for a full week is a systemic failure. For builders, this reinforces the absolute necessity of not just pre-deployment audits but continuous, real-time security monitoring and anomaly detection, especially for custom-modified bridge contracts which remain a primary vector for catastrophic exploits.
One of Ethereum's most profitable and well-known MEV (Maximal Extractable Value) bots, 'jaredfromsubway.eth,' was exploited for approximately $7.5 million on Saturday. An attacker tricked the bot by creating a malicious token and liquidity pool. The exploit didn't rely on a smart contract flaw but on manipulating the bot's own automated logic to grant a lasting token spending approval, which the attacker then used to drain the bot's funds.
Why it matters
This isn't just another DeFi hack; it's a new level of sophistication targeting the automated agents that are now central to on-chain liquidity. The exploit vector—weaponizing token approvals against a bot's own profit-seeking logic—is a stark warning. As MEV and agentic trading become more widespread, securing the operational logic and permissioning systems of these bots becomes as critical as auditing the underlying smart contracts.
The exodus from LayerZero to Chainlink CCIP we've been tracking is accelerating. Following Kraken's public redlining of LayerZero and Turtle's $5.5B migration, Solv Protocol is now moving its entire $700 million tokenized Bitcoin infrastructure to Chainlink. Like the others, Solv cited security concerns stemming from the $292 million KelpDAO exploit.
Why it matters
When $700 million moves, the market listens. This confirms that Turtle's $5.5B migration wasn't an isolated incident, but the start of a broader protocol consolidation. The perceived security and decentralization of the underlying cross-chain infrastructure are now non-negotiable for high-value protocols, and the KelpDAO fallout is actively reshaping the interoperability hierarchy.
The crypto industry's lobbying push to get the CLARITY Act to a Senate floor vote before the August recess just got massive reinforcements. A coalition of over 1,200 mainstream tech companies, including Amazon and Google, has joined crypto leaders like a16z and Coinbase to publicly urge the Senate to pass the bill. The joint letter specifically advocates for the controversial Section 604 developer protections we've been tracking, aiming to shield non-custodial software contributors from liability.
Why it matters
This is a full-court press from both mainstream tech and crypto to end the 'regulation by enforcement' era in the U.S. Passage of the CLARITY Act would be a game-changer, providing the legal certainty needed to build and invest in blockchain-based projects without the constant threat of ambiguous legal action. For builders, this is the most critical piece of pending legislation; its outcome will determine whether the U.S. remains a viable place to innovate in this space.
The FDIC and Treasury are moving to place stablecoin issuers under the same stringent anti-money laundering (BSA/AML) and sanctions (OFAC) compliance regimes as traditional banks. Following the passage of the GENIUS Act, the FDIC published a Notice of Proposed Rulemaking on June 5, opening a 60-day comment period on rules that would apply to 'permitted payment stablecoin issuers' (PPSIs).
Why it matters
This is the formal end of stablecoins operating in a regulatory gray area. The compliance overhead and cost for issuers will now be on par with a chartered bank, which will likely favor large, well-capitalized players like Circle and drive consolidation. For startups building on stablecoin rails, this means any protocol interacting with these issuers will need to have its own compliance story straight, as the requirements will flow downstream.
The on-chain market for tokenized US Treasuries has exploded, crossing $7 billion in value—a 600% increase since January 2025. This growth is being fueled by institutional demand and the new regulatory environment created by the GENIUS Act, which prohibits yield on payment stablecoins. This has pushed stablecoin issuers and other large players to seek yield-bearing, compliant, on-chain Treasury exposure. In response, giants like Fidelity and State Street have launched dedicated stablecoin reserve money market funds this month.
Why it matters
This isn't just about another crypto asset growing; it's the institutional-grade plumbing for a new financial system being built in real-time. The surge in tokenized treasuries provides a compliant, on-chain, yield-bearing alternative for the massive reserves backing the stablecoin economy. It's a critical bridge between DeFi and TradFi, and its growth is a leading indicator of where serious institutional capital is flowing.
After a staggering Q1 where AI startups captured 80% of a record $300 billion in global venture funding, the market is showing the first signs of rebalancing. For the week of June 14-20, AI's share of weekly VC deployment finally dropped below 50%. While AI still pulled in a massive $7.4 billion, significant capital is now flowing back into non-AI deep tech sectors like biotech and quantum computing.
Why it matters
This could be the first indicator that the venture market is starting to look beyond the generative AI hype bubble for returns. For startups outside the immediate AI space—who have watched capital crowd them out for the past quarter—this suggests a slight opening of the funding spigots.
In a delightful fusion of high-stakes racing and pure chaos, Colonial Downs Racetrack will feature corgi races on the opening day of its 11-week thoroughbred summer meet on June 27. The move is part of an effort to attract new and diverse audiences to the traditional sport.
Why it matters
The corgi charm offensive we've been tracking continues. From Pancake taking over as a WNBA mascot to now infiltrating thoroughbred racing, the breed serves as a consistent source of whimsical, highly shareable moments that even traditional institutions are leveraging for community engagement.
DeFi's Security Reckoning A series of costly exploits—a $4.7M infinite-mint bug on the Secret/Axelar bridge and a $7.5M honeypot trap draining a major MEV bot—are exposing critical vulnerabilities in cross-chain infrastructure and automated trading logic. The fallout is driving a flight to quality, evidenced by Solv Protocol's $700M migration from LayerZero to the more battle-tested Chainlink CCIP.
The Shift to 'Loop Engineering' A consensus is forming around 'loop engineering' as the default pattern for agentic coding. Instead of one-shot prompts, developers are building autonomous systems that plan, execute, verify, and retry. This paradigm shift, championed by figures from Anthropic and reflected in new tools, treats AI not as a copilot but as an autonomous production contributor.
Regulators Solidify Their Stance The grace period for crypto regulation appears to be closing. Major tech companies are now lobbying hard for the CLARITY Act to pass, seeking a stable federal framework. Concurrently, the EU is moving on to 'MiCA 2.0' to tackle DeFi, and the US is applying bank-style AML rules to stablecoin issuers under the GENIUS Act, signaling a move toward clear, enforceable, and stricter rules.
L2s Specialize to Compete As the Layer 2 space matures, leading chains are differentiating with architectural upgrades. Base's upcoming Beryl hard fork introduces a native B20 token standard to court institutional RWA and stablecoin business. Meanwhile, Polygon is hitting 5,000 TPS to compete in high-volume payments, showing that L2s are moving beyond just scaling Ethereum to building specialized, defensible ecosystems.
The AI Memory Layer Emerges A new frontier in AI is the development of persistent, adaptive memory. Google's research into RNNs with 'growing memory' and Perplexity's 'Brain' system point toward models that learn and compound intelligence from experience, rather than starting fresh with each prompt. This could mark the next major architectural shift beyond the current transformer-dominated era.
What to Expect
2026-06-25—Base's Beryl hard fork is scheduled to activate on mainnet, introducing the B20 native token standard.
2026-06-27—Corgi races will be featured at the Colonial Downs Racetrack's summer meet opening day.
How We Built This Briefing
Every story, researched.
Every story verified across multiple sources before publication.
🔍
Scanned
Across multiple search engines and news databases
358
📖
Read in full
Every article opened, read, and evaluated
158
⭐
Published today
Ranked by importance and verified across sources
14
— The Chain Reactor
🎙 Listen as a podcast
Subscribe in your favorite podcast app to get each new briefing delivered automatically as audio.
Apple Podcasts
Library tab → ••• menu → Follow a Show by URL → paste