Trezor has warned that a third-party security breach enabled phishing emails to be sent out from the hardware wallet provider’s official domain.

The breach comes soon after a security incident at ShipMonk compromised the personal information of Trezor users.

Trezor Warns Of Third-Party Breach

Trezor issued a warning in an official X post, informing users that the email “Critical Security Alert: STM32 Entropy Vulnerability” was a phishing attempt and urged them to avoid clicking any links.

“Please be aware that the email named ‘Critical Security Alert: STM32 Entropy Vulnerability’ is not coming from us, and it’s a phishing attempt. Do not click on any link.”

The compromised domain has since been taken down, and Trezor has launched a full investigation into the breach and how hackers used the company’s official domain to send phishing emails. Marcello Paz, a crypto commentator, said he received the phishing email in question and shared screenshots asking customers to update their hardware wallets due to a “critical vulnerability.” Unlike typical phishing emails, the email’s credentials showed official domain names and signatures.

“Hello @trezor, I received a “Critical Security Alert: STM32 Entropy Vulnerability” email today (9 Sep 2026). Gmail shows From: Trezor Security , Return-Path: noreply@mailing.trezor.io, Sendinblue campaign, DKIM/SPF/DMARC pass for trezor.io. Body claims a factory STM32 RNG defect (~1 in 4 devices), ~40-bit seeds, and a “check if you’re affected” link via r.mailing.trezor.io plus xPub verification. This matches the entropy-phishing wave, not any official advisory.”

Similar Attempt On BitBox

BitBox, a Swiss Bitcoin hardware wallet maker, reported a similar phishing attempt. The company shared a similar email on its official X account, warning users it was a phishing attempt and urged them to be cautious.

“There is currently a phishing email going around that’s pretending to come from us. Please do not follow the instructions in the email! We are currently investigating.”

Previous Security Incidents

Last month, Trezor’s shipping provider ShipMonk was hit by a major security breach that exposed personal information linked to its customers. Trezor initially disclosed that personal information, including names, cities, and email addresses of 13,700 users, was compromised. However, it said another 67,000 US-based users were affected by the breach.

Hardware wallets have been hit by several security vulnerabilities and breaches recently. Ledger’s security team disclosed a major vulnerability in Trezor Safe 7’s TROPIC01 chip, demonstrating how a lab-based laser attack bypassed its firmware verification system. Ledger suffered a major security breach in 2020 that exposed the personal information of over 270,000 customers, including names, email addresses, phone numbers, and even home addresses. The details were published on a dark web forum, with impacted customers receiving scam calls and physical letters even years later.

Disclaimer: This article is provided for informational purposes only. It is not offered or intended to be used as legal, tax, investment, financial, or other advice.

WROCŁAW, Poland — 11th September 2026 — Getting to the top of Google no longer means getting found by investors. That disconnect is the starting point of HackGPT, the new AI search playbook published by ICODA, a crypto marketing agency founded in 2017 and trusted by more than 650 clients across DeFi, GameFi, iGaming, Exchanges, and Token Sales. The 51-page guide documents how Web3 projects build citation authority inside ChatGPT, Perplexity, and Google AI Overviews — and why teams still treating ChatGPT SEO as an extension of Google strategy are losing ground fast.

The Data Point Every Crypto Team Got Wrong

Only 43.2% of pages ranking #1 on Google are ever cited by ChatGPT — the majority of top-ranked pages are AI-invisible.

That finding, drawn from Semrush’s study of 230,000 prompts, is the central tension HackGPT was built to resolve. For years, Web3 marketing agency strategy treated Google rankings as the default proxy for investor discovery: earn backlinks, publish content clusters, reach Page 1. That logic has fractured.

Semrush’s parallel 1,094-category analysis sharpens the picture: the brand that dominated a topic had more organic traffic than its closest competitor only 48.4% of the time — worse than a coin flip. Organic authority and AI citation authority are measurably different things. Winning one does not guarantee winning the other, and for crypto projects operating in 90-day launch cycles, that gap is a direct revenue risk.

ChatGPT’s citation behavior makes the divergence sharper still. While Perplexity cites Google’s top-10 results 91% of the time, ChatGPT overlaps with those same results just 14% of the time — actively sourcing fresher, more conversational content that standard SEO strategies never touch. For any Web3 marketing agency optimizing exclusively for Google, their clients are invisible on the AI platform now processing over one billion web searches per week.

The data aggregator myth: ICODA and Semrush’s most disruptive finding

CoinGecko, CoinMarketCap, and DeFiLlama earned zero AI citations across 100+ controlled queries — a result that overturns the most repeated GEO advice in crypto.

ICODA and Semrush jointly ran more than 100 controlled query sessions across ChatGPT and Perplexity, covering four categories: recommendation-style prompts, breaking news queries, safety and legitimacy checks, and community-framed questions. Every citation was captured. The three most trusted data aggregators in crypto — CoinGecko, CoinMarketCap, DeFiLlama — did not appear once across any category or engine.

What did get cited consistently: roundup content, financial media placements, the project’s own site, and community platforms. That result forms the structural core of HackGPT. Generative engine optimization for Web3 isn’t about listing your token everywhere investors look. It’s about building the content infrastructure language models actually retrieve.

HackGPT: Inside the 51-page framework

HackGPT converts AI search theory into an execution roadmap built specifically for crypto — not adapted from a generalist SEO playbook.

Who it’s for: SEO leads, CMOs, and growth teams at DeFi protocols, crypto exchanges, token launches, and iGaming platforms who need measurable AI citation gains — not another content calendar.

The framework maps the full citation lifecycle across ChatGPT, Perplexity, Gemini, and Google AI Overviews, with each tactic tied to platform-specific retrieval behavior. Here’s what the 51 pages cover:

• Citation mechanics: How language models retrieve and weight sources — and why most crypto content fails the retrieval test before it ever reaches an investor
• Platform-specific playbooks: ChatGPT and Perplexity require different content architectures; the guide maps both, plus AI Overviews and Gemini
• Documented results: ICODA made ChatGPT its number-one lead source in four months; a Perplexity optimization effort produced a 286% traffic surge with 779 AI-generated sessions
• New performance KPIs: AI Share of Voice, LLM citation tracking by platform, and conversion segmentation by engine — with a monitoring cadence built for the 70% answer variability rate

The yield math behind the shift is clear: AI-referred traffic converts 23 times better than standard organic. For any crypto marketing agency still measuring success by Google rankings, that figure changes the entire budget allocation conversation.

Why crypto-native agencies lead the generative engine optimization race

Crypto-native Web3 agencies adapt faster to AI citation dynamics because they already understand the trust signals language models apply to blockchain claims.

Generative engine optimization in crypto isn’t only a content challenge. Language models apply stricter evaluation criteria to financial claims — weighting source authority, independent corroboration, and regulatory framing far more heavily than they do for general information queries. A traditional agency building GEO strategy for a DeFi protocol without understanding tokenomics language, MiCA compliance framing, or the credibility signals that distinguish trusted crypto sources from promotional ones is building on the wrong foundation.

ICODA’s 14+ years of crypto-native practice translate directly into GEO execution: which publication ecosystems AI engines weight as authoritative for blockchain topics, how investors phrase their research queries, and which community platforms generate the conversational signals that improve AI search visibility for financial products.

Building the citation architecture: Four layers that actually work

AI search visibility for crypto projects is built across four interconnected layers — and most projects invest in only one.

• Financial media placements: Articles in crypto-native publications that AI engines index as authoritative sources carry higher citation weight than aggregator listings across both ChatGPT and Perplexity. HackGPT maps which outlets by query category.
• Roundup and comparison content: Recommendation-style formats — “best of,” comparisons, ranked lists — are the single most commonly cited content type in crypto AI search responses. The format built for synthesis is the format AI retrieves.
• Community platform presence: Forum threads, crypto subreddits, and Discord discussions generate the conversational citation signals that make ChatGPT treat a project as community-verified rather than brand-promoted.
• Owned content structured for AI parseability: Short paragraphs, direct answers in the opening sentence, and clear heading hierarchies all improve how ChatGPT SEO actually functions at the technical level — the layer most crypto content currently ignores.

The citation channel map

Different AI engines weight different source types — a single-channel strategy captures less than 20% of available citation opportunity.

HackGPT prioritizes these channels by project vertical. DeFi protocols, exchanges, and token sales face different citation competitive landscapes — the framework maps execution by project type, not just content format.

The window is open — and closing

The agencies building AI search visibility now will own the citation landscape before competitors recognize that the terrain has changed.

Gartner projects traditional organic search volume will decline 25% by 2026. AI Overview coverage has expanded from 6.5% to 48% of Google queries inside twelve months, and current trajectory points to 70–80% coverage by year-end. In 53.7% of ChatGPT topic categories tracked by Semrush, no brand dominates yet — that blank space is available, but only temporarily.

Research-backed generative engine optimization can lift AI visibility by up to 40%. ICODA’s own results demonstrate that ChatGPT SEO built on citation architecture — not keyword density — can establish an AI engine as a project’s primary lead source within a single quarter. LLM-referred investors convert at 15.9% on ChatGPT versus 1.76% for Google organic. The audience asking AI assistants which protocol to trust is smaller than Google’s total addressable market, and it is decisively higher-intent.

The competitive question for crypto has shifted from who ranks on Google to who gets cited when an investor asks an AI assistant which project to trust. HackGPT is the answer to the second question.

→ Get the HackGPT guide

The CEO of an Ant-backed humanoid robotics startup has published a letter to OpenAI in Chinese that raises questions about technical and design similarities between the two companies’ recent models.

“People often say major tech companies have intelligence networks monitoring the whole internet, this time I believe it, this is a direct distillation of us without any modifications,” Guo Renjie, CEO of Suzhou-based JoyIn, said in a public statement Thursday, according to a CNBC translation.

He said the startup had publicly presented its “extraterrestrial visitor” AI model framework in Silicon Valley a few weeks ago, before OpenAI’s Chief Scientist published “An Alien Mind” on Sept. 6. Guo questioned similarities in core technological approaches such as “recursive self-improvement” and the use of AI to optimize computing power.

Guo also highlighted similarities in the outer space-inspired design of OpenAI’s GPT-6 Astra web page and the website for JoyIn’s Aether model that he claimed was released two months ago. He added the startup has started the process of filing a lawsuit.

CNBC was unable to independently verify the claims. Some concepts are existing parts of AI research more broadly, although companies connect and implement them differently. OpenAI did not immediately respond to a request for comment.

U.S.-based Anthropic has repeatedly flagged unauthorized “distillation” of its models by Chinese companies to improve their own AI capabilities. On Tuesday, a U.S. cybersecurity agency said six Chinese companies, including DeepSeek and Alibaba, distilled models from Anthropic, Google and OpenAI.

JoyIn’s Aether model, which Guo said he decided to publicly announce Thursday, claims its perceptive, rather than text-based, approach to robotic control enables humanoids to complete tasks with a 90% success rate on first attempt.

Zhu Mingxuan, who led the model’s development, said she left U.S. humanoid company Figure last year, where she had also worked on models for helping humanoid robots mimic human actions. She told CNBC earlier this week that Aether reduced training time by two-thirds, and that she planned to open-source parts of the model, such as those relating to touch, but not portions related to energy use.

Getting humanoids and AI models to perform as intelligently as humans across a variety of tasks has remained a challenge, amid a broader tech race between U.S. and Chinese companies.

Garry Tan, chief executive of famed startup accelerator Y Combinator, told CNBC this month that he would “do nothing” about distillation and joined others in pointing out that the U.S. companies have trained their AI models on data covered under copyright law.

—CNBC’s Kate Rooney and Isabel O’Brien contributed to this report.

Brevo, an email delivery platform used across the crypto industry, disclosed that an attacker leveraged a login-system weakness to gain access to multiple client accounts. The incident allowed phishing messages to be sent to a combined audience of roughly 347,000 Trezor newsletter subscribers, with additional campaigns also reaching audiences tied to BitBox and CoinTracking.

In a Thursday postmortem, Brevo said the attacker used six accounts to send phishing emails. It also reported that contacts were exported from 43 accounts, while 93 accounts showed no meaningful activity—though Brevo did not clarify whether those categories overlap. Brevo added that the access-control boundary that should have limited the attacker’s reach to a single organization failed.

Key takeaways

• Brevo reported that an authorization boundary failed after an attacker configured an account with single sign-on and invited real users into the setup.
• At least six Brevo accounts were used to send phishing emails.
• Trezor says the initial phishing email was sent to about 347,000 newsletter customers, and it is treating those addresses as potentially exposed.
• BitBox and CoinTracking also confirmed unauthorized newsletter activity routed through Brevo, though they reported no evidence of lost funds or exposed recovery phrases.

How Brevo’s login flaw enabled cross-account access

Brevo’s postmortem describes a pathway in which an attacker created a Brevo account, turned on single sign-on, and then invited legitimate Brevo users into the configuration. Brevo said the design should have confined access to the organization associated with the configuration, but the authorization boundary did not hold.

As a result, the attacker was able to reach every organization the invited users could access. Brevo’s write-up links the exposure directly to this breakdown in access controls, rather than to a breach of the affected organizations’ own systems.

The incident surfaced publicly after warnings from Trezor and BitBox earlier in the week, which pointed to their shared email provider and explained why the fraudulent emails appeared credible and passed ordinary authentication checks.

Phishing mechanics: what recipients were asked to do

Trezor said the phishing email—titled “Critical Security Alert: STM32 Entropy Vulnerability”—included a link to an app designed to solicit wallet backups. According to Trezor, the company disabled the domain at the DNS level within about 20 minutes. Even with the rapid takedown, Trezor reported that about 2,500 people accessed the link before it was blocked.

Trezor also emphasized the broader risk to its subscriber list. In comments provided to Cointelegraph, a Trezor spokesperson said the initial email was sent to 347,000 customers, and that all recipients were subsequently contacted about the danger.

The spokesperson added: “Until we hear more from Brevo, we are treating all roughly 347,000 newsletter addresses as known to the attacker and possibly reusable for phishing.” Trezor further stated that its Brevo account stored only opt-in newsletter email addresses and no other customer data.

Hardware wallet and crypto services respond: exposure without confirmed credential theft

BitBox told Cointelegraph that its unauthorized email was delivered through Brevo and appeared to reach its full newsletter and tutorial audience.

In its response, BitBox said Brevo held only email addresses and language preferences for it. BitBox reported no evidence of compromised company credentials, no indication that attackers downloaded data beyond the newsletter contacts, and no signs of funds being stolen or recovery phrases disclosed. Still, it said it is treating the list as potentially accessed while awaiting Brevo’s logs.

CoinTracking, meanwhile, reported separate phishing activity. The company said its Brevo account distributed an email titled “Data Breach Notice: Please refresh API Keys as soon as possible.” CoinTracking warned recipients not to click the links in the message, indicating that the main threat was credential-related phishing rather than immediate compromise of underlying systems.

Together, the responses underline a common pattern in third-party email incidents: the most immediate harm may be messaging-based, but the bigger operational concern is whether contact lists can be reused for follow-on attacks.

What Brevo disclosed—and what remains unclear

Brevo’s incident report focuses on the account-access path, but some details remain ambiguous for downstream victims. Brevo said contact exports occurred across 43 accounts and that 93 accounts showed no meaningful activity, without specifying whether those numbers overlap or how many organizations were fully affected end-to-end.

Brevo also did not provide, in the disclosed summary, a precise mapping from the six sending accounts to the different affected crypto companies’ audiences. Cointelegraph attempted to request additional information from Brevo but received no response before publication.

For investors, traders, and builders, the relevance extends beyond the immediate phishing harm: reputable crypto firms rely on email service providers to communicate security alerts, product updates, and documentation. When those communications channels can be abused—especially when phishing content looks authentic—users may face repeated attempts that target them again using addresses already in the attacker’s possession.

Going forward, recipients of such newsletters should be cautious about any unexpected security prompts, verify warnings through official channels, and avoid entering sensitive data into links from unsolicited messages. The core uncertainty now is how thoroughly Brevo’s investigation identifies which organizations’ contacts were exported versus merely accessed, and whether the attacker obtained broader metadata that could support additional phishing campaigns.

Crypto firms and their customers should watch for follow-on updates from Brevo’s incident findings—particularly any clarification on which accounts were used for exports and whether any categories of access overlap—while continuing to educate users to treat “urgent security alerts” sent via newsletter channels as untrusted until verified independently.