supply chain

Bilingual

Scanning 7.6 Petabytes of HuggingFace Training Data for Secrets
2 months ago
- A scan of 7.6 petabytes of public AI training data on Hugging Face found 221,303 live, unique credentials in 6,003 datasets.
- The highest-impact secret exposed 393 GB of PII, covering an estimated 3.7% of the global population.
- Supply chain risks include 349 live GitHub personal access tokens (223 with full repo write, 130 with CI workflow write, 112 with admin:org, 110 with package publishing) and 318 Docker Hub tokens that can push images.
- Live cloud credentials include 8,557 GCP service-account keys, 3,343 AWS keys passing STS identity checks, and 8,594 live database logins.
- 5,885 live Slack tokens and Mailgun keys were found, with examples from a Fortune 500 tech workspace and major brand domains.
- A live AWS key from a Brazilian fintech leaked via a chatbot conversation, captured and mirrored 18 times in the training data.
- At least 11,496 live AI provider keys (OpenAI, Azure, Anthropic, etc.) were found, with a potential minimum cost of $920,000 per year in stolen inference.
- Hugging Face tokens include 787 live ones (237 with write access, 70 with org-admin), with 63 self-leaks and 700 scraped from elsewhere.
- Leaks multiply: 44% of secrets appear in more than one dataset, with one key copied into 1,131 datasets.
- Named open models and corpora contain thousands of live keys, including StarCoder, OLMo 3, SmolLM2, and Huginn.
- Recommendations: scan before publishing, scan before training, rotate any key that touched a public surface, and enable leak alerts.
- Training data leaks are permanent; only rotation (revocation) neutralizes them.
- The scan was conducted with responsible disclosure to Hugging Face, who are also contributing storage bucket scanning support to TruffleHog.
Tesla sues Cybertruck supplier Angstrom amid standoff keeping production hostage
2 months ago
- Tesla is seeking emergency court access to a supplier's plant in Troy, Texas, to retrieve tooling needed for Cybertruck production.
- The supplier, Angstrom Automotive Group, is closing the plant and has denied Tesla entry, jeopardizing thousands of Cybertruck orders.
- Tesla's filing is narrow, requesting only retrieval of its property, not litigation of other disputes.
- The specialized tooling (die-cast machinery, trim dies, etc.) is custom-built for Cybertruck parts and cannot be replaced quickly.
- Angstrom acquired Anderton Castings, the original supplier, and inherited Tesla's contract; the plant closure followed a demand for weekly payments.
- Tesla's on-hand parts are nearly exhausted, and no other supplier can currently make these components, leading to a production halt.
- The article notes potential supplier tensions due to Cybertruck's lower-than-expected sales (20,000/yr vs 250,000/yr planned).
Reading List 06/20/26
2 months ago
- The US House and Senate reached a deal on the combined ROAD to Housing Act, eliminating burdensome build-to-rent provisions.
- Housing developments with grocery anchors like Costco or Safeway may face less NIMBY opposition due to trusted retailer co-sponsorship.
- The Squamish Nation in Canada built the massive Senakw housing project on restored land, free from Vancouver's restrictive planning rules, expected to generate C$10 billion.
- Phoenix Semiconductor is recreating obsolete microchips for long-life military equipment like F/A-18 Super Hornet jets.
- The Trump Administration invoked the Defense Production Act to address supply constraints in munitions manufacturing.
- China's export restrictions on indium phosphide (InP), a critical material for AI data center photonics, are causing supply bottlenecks in the US.
- Two Michigan lawmakers proposed a bill to ban Chinese cars from entering the US, adding to existing EV tariffs.
- The IEA released its 2026 electric vehicle trends report, and RAND published a report on China's industrial policy evolution.
- The US is concerned that China may possess one of ASML's most advanced EUV lithography machines.
A breakthrough in C/C++ dependency management
2 months ago
- A new technology allows remote, on-demand includes in GCC and clang via #include <URL>.
- The tool is provided as a package (remote_includes.tgz) that must be compiled and used with LD_PRELOAD.
- The author warns heavily against using this technology, stating it is dangerous and will cause harm.
- The functionality enables running code directly from URLs, bypassing local file security.
- The post is satirical, mocking supply chain vulnerabilities and the ease of remote code execution.
February Shenzhen Trip Update
2 months ago
- Traveled to Shenzhen to meet 16 vendors and suppliers over 5 days, but got sick on day 6 and canceled meetings.
- Successfully tested the first version (EVT1) of the new watch circuit board, achieving under 100uA average power draw.
- Open source PebbleOS development progressed with 92 new commits; a colleague integrated the nimBLE Bluetooth stack for potential support on existing Pebble watches.
- Shenzhen's supply chain offers unmatched speed and access to components, with business decisions often made within minutes via WeChat.
- In-person meetings are crucial for idea generation, problem-solving, and building trust, despite the convenience of digital communication.
- Now working with wearable ODMs instead of OEMs, as the smartwatch industry has matured since Pebble's early days.
- WeChat's translation feature and acceptance of US credit cards via WeChat Pay/AliPay have greatly simplified travel and communication.
A quote from Seth Larson
2 months ago
- PyPI now blocks new file uploads to releases older than 14 days.
- This restriction prevents poisoning of old and stable releases in case of compromised tokens or workflows.
- The abuse hasn't been observed yet, but was technically possible without this measure.
- The change was announced by Seth Larson on the PyPI blog on 23rd July 2026.
RubyGems Supply Chain Attack
2 months ago
- A supply chain attack on RubyGems was discovered via a new gem called git_credential_manager that downloaded malicious binaries from a Forgejo instance.
- The gem evolved over four versions in nine hours, starting as a dropper, then silencing output, and finally executing on library load rather than explicit command.
- The malware avoids CI environments by checking for over 30 environment variables, targeting developer laptops.
- The attacker also compromised other gems (Dendreo and fastlane-plugin-run_tests_firebase_testlab) by adding the malicious gem as a dependency or hijacking accounts.
- Two long-dormant maintainer accounts were reactivated to push the malicious code, highlighting the risk of inactive accounts in package registries.
- This marks RubyGems' first major incident of this type, similar to attacks previously seen on npm and PyPI.
New Research: A "Verified" GitHub Commit Is Not Unique
2 months ago
- Git uses content hashes to uniquely identify commits, which supply-chain tools rely on as immutable names for signed content.
- A preprint reveals hash chain malleability: an attacker can create a different signed commit with the same tree, metadata, and a valid signature, changing only the hash and affecting subsequent commits.
- The issue stems from signature malleability in GPG-based schemes (ECDSA, RSA, EdDSA) and S/MIME, not from hash function weaknesses like SHA-1 collisions.
- GitHub's server-side verification accepts these malleable signatures without canonicalization, issuing 'Verified' badges for each variant, and retains verification even if keys are later revoked.
- Local tools like git verify-commit show mixed behavior, accepting some malleable signatures but rejecting others that GitHub accepts.
Micron CEO says low-price push by customers fueled the memory shortage
3 months ago
- Memory and RAM shortage is driving up smartphone and computer prices, partly due to aggressive price pressure from customers causing reduced manufacturer investment.
- Micron CEO notes that prices dropped to one-third of previous levels in 2023, leading to poor margins and halted industry investments, with shortages potentially lasting beyond 2027.
- AI companies' increased demand for memory chips exacerbates the supply-demand imbalance, resulting in consumer price hikes, including Apple's recent increases on Mac and iPad models.

supply chain

Scanning 7.6 Petabytes of HuggingFace Training Data for Secrets

Tesla sues Cybertruck supplier Angstrom amid standoff keeping production hostage

Reading List 06/20/26

A breakthrough in C/C++ dependency management

February Shenzhen Trip Update

A quote from Seth Larson

RubyGems Supply Chain Attack

New Research: A "Verified" GitHub Commit Is Not Unique

Micron CEO says low-price push by customers fueled the memory shortage