Invest1 distinct publisher3 min readUpdated
Rapid7's Operation ASTERIX report shows the cost of building convincing wallet malware collapsing while targeting stayed manual. The durable asset is the validated list, not the code.
The Investor · Invest desk
Compiled by The InvestorSomething wrong?How this is made
Rapid7 Labs published on August 17 the anatomy of a live crypto phishing pipeline it calls Operation ASTERIX, found by researchers Anna Širokova and Jan Recinsky after the operators left a web directory exposed on their own infrastructure [1][2]. Two details matter for anyone holding customer records: the crew used GitHub Copilot across the entire build of counterfeit Ledger, Trezor and Exodus apps, and it had already confirmed that 43,066 phone numbers on a leaked list belonged to real crypto exchange users [3][4].
The open directory held roughly 885,000 phone numbers, of which the largest single file was 316,002 German mobile numbers [5][6]. Rapid7 says the operators ran those German numbers through an account checker and confirmed 43,066 as exchange users, a hit rate of about 13.6 percent [7]. Read that denominator carefully. The headline summary frames the 43,066 against the full 885,000 [4], but the report's own detail attributes the checks to the German file, which is about 36 percent of the total set [1]. Smaller directories covered Hong Kong, Bulgaria, the UK, the US, Canadian fintech customers and Ledger-related lists [8]. A further 5,576 numbers tied to Binance accounts were queued for attack, roughly one in eight of the validated pool [9][2], and the server also held a Kraken checker and fake emails posing as Crypto.com [10].
The payload is unglamorous and effective. The apps imitate Trezor Suite and Ledger Live, with Exodus also spoofed, and ask the user to type a 12 to 24 word recovery phrase [11]. That phrase is the master key to the wallet, and it was exfiltrated over Telegram [12][13]. Recovered prompts, shell history and project files show AI assistants used to package the Electron apps, obfuscate code, fix builds and prepare the malware for distribution, not merely to produce snippets [14][15]. When one model started refusing parts of the work, Rapid7 says the operator switched providers and tried a custom jailbreak prompt on the next one [16].
Then the activity logs deflate the picture. They record only 20 lead lookups across about two weeks and six phishing emails sent, which Rapid7 reads as slow hand-picked targeting rather than mass contact [17][18]. At 20 lookups per fortnight, working through 43,066 validated targets would take on the order of 80 years [3]. So the assistant compressed engineering, not conversion. The bottleneck is the human on the Asterisk telephony rig recovered from the server, placing voice-phishing calls timed to land alongside fake support emails the victim had already received [19][20].
That reorders the risk. The scarce input is not code, it is a list of confirmed self-custody customers with a phone number attached, and that list comes from your vendors. Earlier in August, Trezor warned 13,689 customers after a breach at shipping partner ShipMonk exposed names, emails, phone numbers and addresses [21]. Ledger and Trezor owners have also received physical letters carrying QR codes to phishing sites, according to Cryptopolitan reporting in February [22]. Hacken puts phishing and social engineering at $306 million of the industry's $482 million in first-quarter losses, about 64 percent [23][4].
Watch three things. Whether Copilot-class vendors can detect this build pattern, given the operator's response to refusal was to switch providers [16]. Whether logistics and support partners get treated as wallet-security surface after ShipMonk [21]. And whether the manual bottleneck holds: the campaign was still running when Rapid7 found it, and the firm said it was able to notify providers and authorities including Apple's security team mid-campaign [24][25].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
Rapid7 Labs uncovered Operation ASTERIX, a crypto fraud pipeline that used AI coding assistants to create fake Ledger, Trezor and Exodus apps; the pair's report was dated August 17.
An exposed web directory on campaign infrastructure was discovered by Rapid7 researchers Anna Širokova and Jan Recinsky.
The AI coding tool used was GitHub Copilot, and Rapid7 found AI assistants were used across the entire development process, not just for isolated snippets.
The largest file in the directory was a collection of 316,002 German mobile numbers.
Smaller directories on the server covered Hong Kong, Bulgaria, the UK, the US, Canadian fintech customers and Ledger-related lists.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Detailed forensic account, single-outlet relay
The underlying material is strong in kind: an open-directory seizure with named researchers, artifact inventories, prompt and shell-history recovery, and precise counts (885,000 numbers, 316,002 German, 43,066 validated, 5,576 Binance, 20 lookups, six emails). But the supplied cluster contains only one article relaying Rapid7's report rather than the report itself, and that article's own summary and body disagree on the denominator behind the 13.6 percent hit rate, which caps evidentiary strength.
Live pipeline, human-limited throughput
Real-world use is established rather than hypothetical: tooling was in active use or development when it leaked, the campaign was ongoing at discovery, and 43,066 numbers had been enriched with 5,576 Binance targets queued. Actual exercise of that capability was small, however, with logs showing only 20 lead lookups in about two weeks and six phishing emails sent, and the supplied source reports no victim count, loss figure or install volume for the fake apps.
Scale framing outruns measured activity
The headline and summary lead with corpus scale and AI-built malware ('885,000 phone numbers', 'AI-coded'), and the summary's 43,066-of-885,000 phrasing inflates the validated share relative to the body's German-list arithmetic. The measured operation was small and manual. The gap is moderate rather than severe because the article itself flags that the validated-target count sits oddly against the logs and reports the 20 lookups and six emails plainly.
Vendor threat research relayed by crypto media
The primary findings come from Rapid7, a commercial security vendor whose published research supports its market position, and the supporting loss statistics come from Hacken, another security firm. The relaying outlet is crypto-vertical media that closes with a newsletter solicitation and self-citations to its own earlier phishing coverage. None of these interests is disclosed in the piece, though the specificity of the artifacts and the inclusion of throughput data that undercut the scale framing argue against pure promotion.
Plausible and specific, but uncorroborated
Confidence is held down by single-publisher coverage with no primary document in the cluster, no comment from GitHub/Microsoft, the impersonated wallet vendors, the named exchanges or Apple, and an internal numerical inconsistency. It is held up by the concreteness of the forensic detail, the named researchers, and the fact that the reporting includes evidence against its own strongest framing.
security
Prompts, shell history, a jailbreak: AI assistants turn up inside a crypto-fraud build pipeline1 distinct publisher
invest
The $700 case: Binance's real jurisdictional risk is its subpoena desk, not its vault1 distinct publisher
invest
One $400M round held up crypto's July: 41 rounds and 140 investors, both at lows1 distinct publisher
invest
Lido's buyback only fires above $40M of revenue. Lido is running about $38M.1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 20, 2026