Connect with us
DAPA Banner

Crypto World

Viral BridgeBench Post Claims Claude Opus 4.6 Was ‘Nerfed,’ Critics Call It Bad Science

Published

on

Viral BridgeBench Post Claims Claude Opus 4.6 Was ‘Nerfed,’ Critics Call It Bad Science

BridgeMind AI claimed Anthropic’s Claude Opus 4.6 was secretly degraded after a hallucination benchmark retest. The viral post has since drawn sharp criticism for flawed methodology.

The claim triggered widespread debate over whether AI companies are quietly downgrading paid models to reduce costs.

BridgeMind Claims a 98% Surge in Hallucinations

BridgeMind, the team behind the BridgeBench coding benchmark, posted that Claude Opus 4.6 had fallen from second to tenth place on its hallucination leaderboard. Accuracy reportedly dropped from 83.3% to 68.3%.

“CLAUDE OPUS 4.6 IS NERFED. BridgeBench just proved it. Last week Claude Opus 4.6 ranked #2 on the Hallucination benchmark with an accuracy of 83.3%. Today Claude Opus 4.6 was retested and it fell to #10 on the leaderboard with an accuracy of only 68.3%,” they wrote.

The post framed this as proof of “reduced reasoning levels.” However, a closer look at the underlying data tells a different story.

Advertisement

Critics Say the Comparison Is Fundamentally Flawed

According to computer scientist Paul Calcraft, the claim is “incredibly bad science,” highlighting a critical problem with the methodology.

“Incredibly bad science You tested Opus on 30 tasks today, previous score was on just *6* tasks Results for 6 tasks in common: 85.4% score today vs. 87.6% prevly. Swing is mostly from a *single* fabrication without repeats – easily statistical noise,” commented Calcraft.

The original high score came from just six benchmark tasks. The new retest expanded the benchmark to 30 tasks.

On the six overlapping tasks, performance was nearly identical, dropping only from 87.6% to 85.4%.

That small swing came mostly from a single extra fabrication in one task. With no repeated runs, this falls well within normal statistical variance for AI models.

Large language models are not deterministic, and one bad output on a small sample can shift results significantly.

Broader Frustrations Fuel the Narrative

Still, the post struck a nerve. Since its February 2026 launch, Claude Opus 4.6 has faced persistent complaints about perceived quality decline.

Developers report shorter responses, weaker instruction-following, and reduced reasoning depth during peak hours.

Advertisement

Some of this traces to deliberate product changes. Anthropic introduced adaptive thinking controls that let the model self-adjust its reasoning budget. The default effort level was later set to medium, prioritizing efficiency over maximum depth.

An independent analysis of over 6,800 Claude Code sessions found reasoning depth dropped roughly 67% by late February.

The model’s file-read ratio before editing code fell from 6.6 to 2.0. That suggests it attempted fixes on code it had barely reviewed.

Advertisement

What This Means for AI Users

This reflects a growing tension in the AI industry. Companies optimize models for cost and scale after launch, while heavy users expect consistent peak performance. The gap between those priorities erodes trust.

Based on the available evidence, the BridgeBench data does not prove a deliberate downgrade. The benchmark comparison was apples-to-oranges, and the overlapping results were nearly identical.

However, the underlying frustration is not entirely baseless. Adaptive compute controls and service-level optimizations have changed how Claude Opus 4.6 behaves in practice. For developers relying on consistent output, those changes matter.

Anthropic has not issued a public statement on the specific BridgeBench claims as of April 13.

Advertisement

The post Viral BridgeBench Post Claims Claude Opus 4.6 Was ‘Nerfed,’ Critics Call It Bad Science appeared first on BeInCrypto.

Source link

Advertisement
Continue Reading
Click to comment

You must be logged in to post a comment Login

Leave a Reply

Crypto World

Circle’s Allaire says USDC freezes require legal orders amid rising criticism

Published

on

Circle (CRCL) may rally another 60% driven by stablecoin adoption, AI agentic finance: Bernstein

Circle Internet (CRCL) CEO Jeremy Allaire offered his clearest public response yet to growing criticism over how the stablecoin issuer handles illicit funds, saying it does not freeze wallets unless there is a formal legal basis to do so.

Speaking on stage at a press conference in Seoul, Allaire positioned USDC, the second-largest dollar-pegged stablecoin, as a regulated financial product rather than a tool for real-time intervention.

“Circle has a very, very clear performance obligation under the law,” Allaire said. “Circle follows the rule of law, and we are able to undertake actions such as freezing a wallet at the direction of law enforcement or the courts.”

Allaire framed USDC as part of the traditional financial system, subject to legal process and oversight. Decisions to blacklist or freeze funds, he suggested, should not be made at the discretion of the company in the heat of an exploit, but instead follow requests from law enforcement or court orders. The approach reflects Circle’s broader strategy to align closely with regulators and institutions.

Advertisement

Rival Tether, the issuer of the world’s largest stablecoin, USDT, has a more proactive approach. The company has repeatedly frozen funds linked to hack and illicit activity within hours. In several cases cited by blockchain sleuth ZachXBT, including exploits affecting Ledger and Remitano, Tether blacklisted stolen funds while equivalent USDC remained untouched.

Allaire’s remarks come at a time of mounting scrutiny. Earlier this month, Drift Protocol suffered a suspected North Korea-linked exploit that resulted in losses of up to $280 million. Roughly $230 million in USDC was moved across chains over several hours. The incident has become a focal point for critics who argue that Circle is failing to act despite having the technical ability to do so.

Intervention carries risks, too

ZachXBT is among the most vocal. In a widely circulated thread on X, he said Circle’s inaction across more than a dozen cases since 2022 has contributed to over $420 million in illicit funds escaping. He pointed to multiple incidents where stolen USDC remained in identifiable wallets for hours or even days without being frozen, including exploits affecting Cetus, SwapNet, and Nomad.

Critics say the pattern highlights a deeper issue. USDC is centrally issued and contains controls that allow Circle to block addresses. Yet those powers are rarely used in real time. By deferring to legal processes that move far more slowly than blockchain transactions, they argue, Circle creates a gap that attackers can exploit.

Advertisement

Others in the industry argue that faster intervention carries its own risks. Omid Malekan, an adjunct professor at Columbia Business School, responded to calls for discretionary freezes by warning that allowing issuers to act beyond legal requirements would undermine the foundations of decentralized finance (DeFi).

Such powers could erode trust in DeFi systems by introducing centralized points of control, Malekan said.

“If Circle and other stablecoin issuers implement arbitrary freeze or seize functions beyond what the law requires, then not only is code not law, but also law is not law,” he wrote on X. “Instead what a single executive inside a single corporation decides is law.”

Source link

Advertisement
Continue Reading

Crypto World

Hyperbridge exploited less than two weeks after April Fools’ day hack prank

Published

on

Hyperbridge exploited less than two weeks after April Fools’ day hack prank

Self-styled “unbreakable” Hyperbridge protocol has been exploited, less than two weeks after making a tasteless April Fools’ joke about being hacked.

Despite previously explaining how a hack was impossible as part of the April 1 prank, the project acknowledged the exploit in a “bridge update!” posted to X. 

According to crypto security firm CertiK, the hacker “forged message to change the admin of Polkadot token contract on Ethereum and profited ~$237K from minting and selling 1B tokens.”

Another on-chain analyst flagged a further 245 ether (worth over $500,000) which was allegedly drained from the project’s TokenGateway contract before being deposited into Tornado Cash.

Advertisement

While this loss may be modest compared to many crypto hacks, especially bridges, many have focused on the karma dealt to a project with a consistently cavalier attitude towards security.

Read more: Bitcoin Depot didn’t spot 50 BTC hack for three days, report

Hyperbridge claimed the North Korean Lazarus Group had drained $37 million on April 1. The announcement linked to a (now deleted) blog post which contained a Rickroll gif before explaining “Why Hyperbridge Can’t Be Hacked.”

Following backlash, Hyperbridge’s “mad scientist,” who goes by “Web3 Philosopher” on X, boasted of the protocol’s “incorruptible” infrastructure.

Advertisement

In February, they also posted screenshots which appear to show correspondence with a big bounty hunter flagging critical vulnerabilities, who was told “exploit them if you found them.”

Apparently taking the April Fools’ prank as a challenge, a known exploiter address began testing Hyperbridge. The attempts were dismissed with “hope you have a quantum computer bro.”

Got a tip? Send us an email securely via Protos Leaks. For more informed news and investigations, follow us on XBluesky, and Google News, or subscribe to our YouTube channel.

Advertisement

Source link

Advertisement
Continue Reading

Crypto World

Foundry’s institutional Zcash pool captures a third of new issuance

Published

on

Cyclops raises $8m for enterprise stablecoin infrastructure

Foundry’s U.S.‑based, compliance‑first Zcash pool has already grown to roughly one‑third of network hashrate, giving institutional miners a regulated way into privacy coins while stoking fresh centralisation fears.

Summary

  • Bitcoin mining giant Foundry has launched an institutional Zcash pool that already accounts for roughly one‑third of new ZEC issuance.
  • The U.S.‑based, compliance‑focused pool is pitched at institutional and public miners as a “purpose‑built” alternative to offshore privacy‑coin infrastructure.
  • Foundry argues Zcash’s zero‑knowledge privacy with selective disclosure makes it more compatible with regulation than rivals like Monero.

Foundry Digital, operator of the Foundry USA Bitcoin mining pool, has officially launched an institutional‑grade Zcash (ZEC) mining pool that has quickly grown to around 30% of the network’s hashrate, consolidating a significant share of new ZEC issuance under a single U.S.‑regulated operator. The Rochester, New York‑based firm, which Fortune notes already commands about 31% of global Bitcoin production, is positioning its new pool as the default home for institutional miners seeking exposure to privacy‑focused assets without abandoning compliance.finance.

In a Business Wire release, Foundry said the Zcash pool has seen “rapid and sustained hashrate growth reaching ~30% of the current Zcash network hashrate” since it was first announced on March 11, with “multiple institutional mining customers already onboarded and contributing hashrate.” The company stressed that the pool is “designed for professional mining organizations and public companies that require a U.S.-based, compliance-ready partner, including KYC verification in line with Foundry’s institutional standards,” mirroring the governance of its Bitcoin operation.

Advertisement

Foundry CEO Mike Colyer framed the move as both a bet on Zcash and a response to unmet institutional demand. “Zcash has matured into an institutional‑grade asset, but the mining infrastructure supporting it hasn’t kept pace,” he said, adding that the new pool is “purpose‑built for the operational and compliance requirements of institutional and public miners.”

A CoinMarketCap summary of the launch notes that the pool will offer know‑your‑customer and anti‑money‑laundering checks, transparent payout calculations, reporting tools and 24/7 technical support, with no minimum hashrate required to join.

Zcash, launched in 2016, relies on zero‑knowledge proofs (zk‑SNARKs) to enable shielded transactions that hide sender, receiver and amount while still allowing selective disclosure to auditors or regulators. Foundry and several commentators have argued that this “privacy with a view key” model is more compatible with institutional compliance than fully opaque systems like Monero, which lack native mechanisms for selective transparency.

Advertisement

At the same time, the arrival of a U.S. pool with roughly one‑third of Zcash’s hashrate raises familiar centralisation questions. Unfolded and other mining trackers have previously highlighted that Foundry USA already coordinates about 30% of Bitcoin’s global hashrate, and Mempool.space data shows the pool averaging more than 340 exahashes per second on Bitcoin alone. Adding a Zcash operation that quickly captures around one‑third of ZEC issuance further concentrates influence over block production in a single corporate group, albeit one that stresses its role in “contribut[ing] to the decentralization of Bitcoin’s hashrate” by anchoring North American capacity.

For Zcash, the trade‑off is stark: institutional capital and hashpower are flowing in through a U.S.‑regulated gateway that validates the project’s positioning as a compliant privacy coin, but at the cost of a more concentrated mining landscape. As regulators in the U.S., EU and Hong Kong tighten their grip on stablecoins, exchanges and tokenized assets — a trend explored in recent crypto.news coverage of HKDAP’s launch, MiCA implementation and the CLARITY Act — Zcash’s bet is that privacy with selective disclosure, plus a mining pool built for auditors rather than cypherpunks, is a price worth paying for long‑term relevance.

Source link

Advertisement
Continue Reading

Crypto World

Bitcoin’s 50% Drawdown ‘Priced In’ Quantum Computing Threat: Bernstein

Published

on

Bitcoin's 50% Drawdown ‘Priced In’ Quantum Computing Threat: Bernstein

Bernstein said Monday that Bitcoin’s selloff has already priced in much of the market’s fear around quantum computing, arguing that the threat is real but still manageable rather than an immediate existential risk.

Bitcoin’s (BTC) near 50% drawdown from its $126,198 all-time high in October 2025 is proof that the market has “priced in” several risks tied to a quantum breakthrough, partly thanks to technological progress on zero-knowledge privacy and quantum-proof cryptography that “counterbalance” the AI and quantum acceleration, Bernstein said in a Monday note shared with Cointelegraph.

The note lands two weeks after Google researchers said future quantum computers could break the elliptic-curve cryptography used across many blockchains with fewer than 500,000 physical qubits in some architectures, reviving debate over how quickly Bitcoin needs a post-quantum upgrade path. This research suggested a quantum computer could crack a Bitcoin private key in nine minutes, in a theoretical scenario, which is less than Bitcoin’s 10-minute block production time.

However, Bernstein said Bitcoin core developers have “adequate time” to determine a post-quantum path. Last week, Bernstein predicted that Bitcoin has about three to five years to prepare for a post-quantum security upgrade, Cointelegraph reported on Wednesday.

Advertisement
Graph showing the risk that an on-spend quantum attack that takes 9 minutes to derive a private key succeeds against Bitcoin. Source: Google Quantum AI

Institutions will play constructive role in quantum-proofing Bitcoin

Bernstein said large institutional holders, including exchange-traded fund (ETF) issuers and corporate treasury buyers such as Strategy, are likely to play a constructive role in any eventual consensus on a post-quantum upgrade.

“We expect institutional partners with now billions at stake to play a constructive role in building consensus on the post-quantum path.”

The note also highlighted the recently introduced BIP-360 proposal and added that slower consensus from Bitcoin developers is seen as responsible behavior when it comes to a $1.5 trillion asset.

BIP-360 is a draft Bitcoin Improvement Proposal that proposes a Pay-to-Merkle-Root output type designed to reduce long-exposure quantum risk by removing Taproot’s key-path vulnerability, though it does not itself add post-quantum digital signatures.

Bernstein said BIP-360 could be implemented as a soft fork for exposed Bitcoin addresses, but added that this would still leave around 8% of the BTC supply in inactive addresses vulnerable to future quantum breakthroughs.

Related: Bitcoiners push for quantum-resistant BIP-360 upgrade as debate heats up

Advertisement

Quantum-proofing Bitcoin is a social issue, not technical

The real challenge of quantum-proofing Bitcoin lies in the societal adoption element of the new standards, not the technical development, according to Arthur Breitman, co-founder of Tezos blockchain.

“The coding work could be done this afternoon,” but Bitcoin holders would still need to migrate to this new standard, Breitman told Cointelegraph during an interview at EthCC 2026.

“If Bitcoin needed to migrate in the next month, they could do it from a technical perspective […] but they can’t get everyone to migrate their key in a month, Breitman said. “It’s going to take years for people to properly migrate their keys,” he added.

Arthur Breitman, co-founder of Tezos, interview at EthCC 2026. Source: Cointelegraph

Asset manager Grayscale’s head of research, Zach Pandl, shared a similar view in a research report last Monday. He said Bitcoin’s quantum-proofing challenges are “more social than technical,” provided that its UTXO model does not have native smart contracts and that some address types are not quantum vulnerable.

However, he warned that the community needs to find consensus on how to quantum-proof wallets where the private key has been lost or is otherwise inaccessible.

Advertisement

Magazine: AI has dramatically accelerated the quantum threat to Bitcoin: AI Eye