Samwise Tech/AI/Robotics Newsletter
Tuesday, June 30, 2026
Agentjacking Attack Hijacks Claude Code via Fake Sentry Errors, Exposing 2,388 Organizations
A new AI coding-agent attack called “agentjacking” achieved an 85% success rate in controlled testing by Tenet Security, which found 2,388 organizations with publicly exposed Sentry credentials vulnerable. The attack sends a crafted error event through a public Sentry API key requiring no breach; Claude Code, Cursor, or Codex retrieves it via the Sentry MCP server and executes attacker instructions using the developer’s full privileges. No credentials are stolen and no policy is violated. Sentry called the flaw “technically not defensible,” and the Cloud Security Alliance classified agentjacking as a systemic MCP vulnerability class within days. VentureBeat reports Datadog, PagerDuty, and Jira connectors share the same exposure.
Sources: VentureBeat Share ↗ ✉︎ Email 💬 Text
Robot Hand Startup Proception Settles Tesla Trade Secret Suit, Raises $11M to Ship Dexterous Hand
Robotics startup Proception settled a trade secret lawsuit with Tesla and announced an $11 million seed round led by First Round Capital, with contributions from Y Combinator and BoxGroup. Tesla had accused founder Jay Li, a former technical lead on the Optimus humanoid program, of stealing secrets related to advanced robotic hand sensors. Tesla dismissed the lawsuit as part of the settlement. Proception is now shipping its first high-dexterity robotic hand — featuring 22 degrees of freedom and multiple joints per finger — to researchers and companies. Rather than requiring a robot in the loop, the device uses a sensor-laden glove to capture human hand interaction data directly.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
South Korea Pledges Over $550B in Semiconductor and AI Data Center Investment to Ease ‘RAMageddon’
Samsung and SK Hynix committed $518 billion to build four new memory fabs in southwestern South Korea, plus $52 billion for a high-bandwidth memory packaging hub, as part of a national plan to ease “RAMageddon” — a worldwide memory chip shortage driven by AI infrastructure demand. A separate $356 billion commitment from Korean tech and energy companies covers AI data centers through 2035. TechCrunch reports the total South Korean commitment exceeds $900 billion when semiconductor and data center pledges are combined. SK Hynix leads the chip expansion while SK Telecom plans to build 15 gigawatts of AI data center capacity across the country.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
Meituan Reveals LongCat-2.0: The 1.6-Trillion-Parameter Model That Secretly Led OpenRouter Rankings
Chinese food delivery company Meituan open-sourced LongCat-2.0, revealing it as the model behind “Owl Alpha,” an anonymous entry that had led global developer charts on OpenRouter for two months. The 1.6-trillion-parameter Mixture-of-Experts system limits active computation to 48 billion parameters per token and supports a native one-million-token context window through a technique called LongCat Sparse Attention. The model is available on GitHub, Hugging Face, and Meituan’s platform under an MIT license, making it commercially viable without restrictions. VentureBeat reports LongCat-2.0 was trained entirely on Chinese chips, a milestone underscoring China’s semiconductor self-sufficiency ambitions.
Sources: VentureBeat Share ↗ ✉︎ Email 💬 Text
DeepSeek Open-Sources DSpark, Cutting LLM Inference Time by Up to 85% with Speculative Decoding
DeepSeek open-sourced DSpark, an inference framework that speeds up large language model responses by up to 85% without altering underlying model outputs. The system deploys a lightweight “scout” that runs ahead of the main model, predicts likely token sequences, and lets the larger model rapidly validate which steps are safe to accept. DeepSeek applied DSpark to its DeepSeek-V4-Flash and V4-Pro models, and independent developer benchmarks showed approximately 60 tokens per second — roughly 2.3 times the non-speculative baseline. The release includes a technical paper, model checkpoints, and DeepSpec, a codebase for training and evaluating speculative decoding systems, published under MIT license on GitHub and Hugging Face.
Sources: VentureBeat Share ↗ ✉︎ Email 💬 Text
TIDAL Bans Monetization of Fully AI-Generated Music Starting July 15
Music streaming service TIDAL announced a policy, effective July 15, 2026, that bars fully AI-generated music from earning royalties or direct-to-fan sales revenue on its platform. TIDAL will use automated tools to identify and tag such tracks with an “AI” badge; tagged content cannot be monetized. The company will also remove AI-generated music that attempts to impersonate an artist. TIDAL described the policy as a “living document” open to revision as the space evolves. TechCrunch notes the move joins similar policies from Spotify, Apple Music, Deezer, and Qobuz addressing the growing volume of AI-generated content on streaming services.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
AI Job Cuts Near 90,000 in 2026, Yet Heavy AI Spenders Are Growing Headcount Faster
New data complicates the prevailing narrative about AI replacing workers: through May 2026, companies announced nearly 90,000 job cuts attributed to AI, yet firms spending heavily on AI are simultaneously growing headcount faster than peers, even in entry-level roles widely predicted to disappear. TechCrunch reports the divergence is sharpening the debate among economists and policymakers who had expected a cleaner signal. The pattern mirrors earlier technology transitions where displacement and creation occurred in parallel but across different sectors and skill levels. Companies citing AI for layoffs include large enterprises in software, finance, and logistics.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
Tech Pulse
Top Frontier Models: Claude Opus 4.8 (88.6%) | GPT-5.5 (88.7%) | GPT-5.4 xHigh (85.2%)
Top Open Source Models: GLM-5.2 (62.1%) | Qwen3.7 Max (60.6%) | MiniMax M3 (59.0%)
Top Small Models (15–50B): LongCat-2.0-Flash (56.8%) | Mistral Medium 3 (54.1%) | Qwen3 32B (51.9%)
Top Edge Models (0–15B): Gemma 3 12B (44.2%) | Phi-4 Mini (41.8%) | Qwen3 8B (40.3%)
AI Leaders: NVIDIA ~$5.2T | Alphabet ~$4.8T | Microsoft ~$3.3T
Robotics Leaders: ABB Ltd ~$194B | Intuitive Surgical ~$144B | Figure AI ~$39B (private est.)
Curated by JD · samwise.agency
Leave a Reply
You must be logged in to post a comment.