Anthropic launches Claude Tag — a team agent that lives inside Slack
A multiplayer Claude agent that sits in Slack channels, takes delegated tasks, and acts proactively for the whole team.
Anthropic launches Claude Tag — a team agent that lives inside Slack
A multiplayer Claude agent that sits in Slack channels, takes delegated tasks, and acts proactively for the whole team.
- Teams tag @Claude into a Slack channel to delegate work; it breaks tasks into stages and executes independently, learning company context over time.
- Admins scope access to sensitive data, tools and memories per channel; an "ambient" mode follows up on forgotten tasks and flags cross-org items.
- Inside Anthropic's own product team, Claude Tag-generated changes account for 65% of code approved and merged.
- Launches as a research preview on Slack for Claude Enterprise and Claude Team customers.
Implications: Anthropic is pushing Claude beyond the terminal into shared team workflows where agents behave like virtual coworkers — squaring up against Microsoft, Glean and Salesforce's Slackbot.
GitHub Copilot app adds bring-your-own-key model providers
The Copilot app can now run agent sessions against your own model providers via a new BYOK setting.
GitHub Copilot app adds bring-your-own-key model providers
The Copilot app can now run agent sessions against your own model providers via a new BYOK setting.
- Supported providers: OpenAI, Azure OpenAI, Microsoft Foundry, Anthropic, LM Studio, Ollama, and any OpenAI-compatible endpoint.
- Configured under Settings → Model Providers with an endpoint and API key, or just a host for LM Studio / Ollama.
- Keys are stored in the local OS keychain and are never read back by the UI.
- Lets teams route inference through enterprise infrastructure for stricter data-boundary compliance; Business/Enterprise plans need CLI enabled in policy.
Implications: BYOK loosens Copilot's model lock-in and positions it for regulated environments that demand self-hosted or private inference.
Claude Code adds sandbox credential blocking and org model restrictions
New controls keep secrets out of sandboxed commands and let organizations restrict which models users can pick.
Claude Code adds sandbox credential blocking and org model restrictions
New controls keep secrets out of sandboxed commands and let organizations restrict which models users can pick.
- v2.1.187 (June 23) added a
sandbox.credentialssetting that blocks sandboxed commands from reading credential files and secret env vars. - Same release added org-configured model restrictions across the picker,
--model,/modelandANTHROPIC_MODEL, showing "restricted by your organization's settings" when blocked. - v2.1.191 (June 24) added
/rewindto resume from before/clear, remembered sandbox network host approvals, and MCP retry logic. - The June 24 build also cut CPU usage during streaming responses by roughly 37%.
Claude arrives in JetBrains IDEs; Copilot adds org agents and CLI steering
GitHub put Claude into public preview as an agent provider for JetBrains alongside new org-level agents and CLI message controls.
Claude arrives in JetBrains IDEs; Copilot adds org agents and CLI steering
GitHub put Claude into public preview as an agent provider for JetBrains alongside new org-level agents and CLI message controls.
- Claude is in public preview for JetBrains IDEs via the Claude Code CLI; the agent currently runs in bypass-permissions mode, auto-approving file edits and tool calls.
- Admins can now publish custom agents at org or enterprise level, auto-available to eligible users in scope.
- Copilot CLI's Send button gained three options during active requests: Add to Queue, Steer with Message, and Stop and Send.
- The cloud agent is now generally available, plus a debug-logs summary view and a
/modelspicker command.
OpenAI and Broadcom unveil "Jalapeño," OpenAI's first custom inference chip
OpenAI revealed its first in-house AI accelerator, co-designed with Broadcom and optimized specifically for large-language-model inference.
OpenAI and Broadcom unveil "Jalapeño," OpenAI's first custom inference chip
OpenAI revealed its first in-house AI accelerator, co-designed with Broadcom and optimized specifically for large-language-model inference.
- Jalapeño is a reticle-sized ASIC architected around OpenAI's inference systems, built to serve models in ChatGPT and other products.
- Co-developed from initial design to manufacturing tape-out in roughly nine months — among the fastest high-performance ASIC cycles ever achieved.
- Early testing shows performance-per-watt substantially better than current state of the art; scaling planned across the following years.
- Framed as part of OpenAI's plan to "build the full stack" rather than rely solely on third-party GPUs.
Implications: A credible custom-silicon path would cut OpenAI's dependence on Nvidia and reshape frontier-lab cost structures.
Anthropic accuses Alibaba of largest-ever AI model "distillation" attack
Anthropic told US officials that Alibaba-affiliated operators used tens of thousands of fake accounts to illicitly extract Claude's most valuable capabilities.
Anthropic accuses Alibaba of largest-ever AI model "distillation" attack
Anthropic told US officials that Alibaba-affiliated operators used tens of thousands of fake accounts to illicitly extract Claude's most valuable capabilities.
- Anthropic alleges a campaign run April 22–June 5, 2026 using nearly 25,000 fraudulent accounts and more than 28.8 million exchanges with Claude.
- The effort targeted Claude's software-engineering and agentic-reasoning capabilities via "distillation" — training a weaker model on a stronger one's outputs.
- Anthropic attributes it to operators affiliated with Alibaba and its Qwen AI lab, disclosed in a June 10 letter to US Senate Banking Committee members.
- The 28.8M exchanges exceed the combined ~16M from three February campaigns Anthropic linked to DeepSeek, Moonshot AI and MiniMax.
Implications: Escalates US–China AI-IP tensions and pressures policymakers toward stricter export and access controls on frontier models.
Nobel laureate John Jumper to leave Google DeepMind for Anthropic
The AlphaFold lead and 2024 Nobel chemistry laureate is departing DeepMind, deepening Google's senior-talent losses.
Nobel laureate John Jumper to leave Google DeepMind for Anthropic
The AlphaFold lead and 2024 Nobel chemistry laureate is departing DeepMind, deepening Google's senior-talent losses.
- Jumper announced he is leaving after nearly nine years at Google DeepMind, where he led the AlphaFold team, to join Anthropic after a break.
- The move follows Transformer co-author Noam Shazeer's reported departure from Google for OpenAI days earlier, compounding retention concerns.
- Bloomberg reported Alphabet stock fell as much as 7.2% intraday on the news, its steepest single-day drop since February 2026.
Agility Robotics to go public via $2.5B SPAC merger with Churchill Capital Corp XI
The Digit maker will list on Nasdaq as "AGLT," raising over $620M to scale production of its next-gen humanoid Digit v5.
Agility Robotics to go public via $2.5B SPAC merger with Churchill Capital Corp XI
The Digit maker will list on Nasdaq as "AGLT," raising over $620M to scale production of its next-gen humanoid Digit v5.
- Merger with Churchill Capital Corp XI values Agility at $2.5B pre-money; combined company expected to trade under ticker "AGLT."
- More than $620M gross proceeds: ~$420M from Churchill's trust (assuming no redemptions) plus a ~$200M PIPE at $10/share led by Foxconn.
- Agility says it has secured more than $300M in multi-year Digit v5 orders (subject to milestones) across 30+ customers.
- Proceeds earmarked to fulfill orders and scale Digit v5 production, physical AI, software, safety systems and manufacturing.
Implications: A second pure-play humanoid maker reaching public markets signals investors now treat industrial humanoids as a fundable, near-commercial category.
Morgan Stanley nearly doubles its 2026 China humanoid shipment forecast to 50,000 units
A second upgrade this year reflects China's rapid shift from trade-show demos to commercial deployments.
Morgan Stanley nearly doubles its 2026 China humanoid shipment forecast to 50,000 units
A second upgrade this year reflects China's rapid shift from trade-show demos to commercial deployments.
- Forecast raised to 50,000 units for 2026, up from a prior 28,000 (itself doubled from a January estimate of 14,000).
- Morgan Stanley sizes China's humanoid market at ~$2B this year, projecting growth to $15B and 446,000 annual shipments by 2030.
- Drivers cited: accelerating commercial validation (including a 6.8B yuan / ~$1.0B order from State Grid), policy support, and aggressive supply-chain expansion.
Implications: China's humanoid sector is entering an early commercialization phase, widening the gap with US makers still in pilot stages.
Bear Robotics to acquire UK humanoid startup Kinisi Robotics
The hospitality-robot maker absorbs Kinisi's KR1 wheeled humanoid and manipulation AI to build an end-to-end physical AI platform.
Bear Robotics to acquire UK humanoid startup Kinisi Robotics
The hospitality-robot maker absorbs Kinisi's KR1 wheeled humanoid and manipulation AI to build an end-to-end physical AI platform.
- Definitive agreement announced; financial terms not disclosed; deal expected to close within days.
- Brings Kinisi's KR1 wheeled humanoid (pick, place, sort, transport), manipulation software, and engineering team to Bear.
- Adds Kinisi's vision-language-action models, robot foundation models and manipulation training data to Bear's deployed-fleet operational data.
- Kinisi founder/CEO Brennand Pierce — a Bear Robotics co-founder — returns as chief robotics officer after closing.
Army to open domestic ranges to vendors, build sites mimicking Ukraine's frontlines
The Army will widen industry access to its test ranges to speed evaluation of counter-drone and interceptor systems under contested battlefield conditions.
Army to open domestic ranges to vendors, build sites mimicking Ukraine's frontlines
The Army will widen industry access to its test ranges to speed evaluation of counter-drone and interceptor systems under contested battlefield conditions.
- Army Secretary Dan Driscoll said companies currently wait 12–18 months to book range time; the service aims to cut that to within 4–6 weeks.
- At least two domestic ranges will be configured to mimic Ukraine's forward lines, with electronic-warfare simulation and contested environments for joint drone/counter-drone testing.
- The Army also plans an overseas range for "much more aggressive testing"; Driscoll declined to name the partner nation.
- Threat context cited: Russia produces ~5,000 Shahed-type drones monthly; Ukraine makes ~30,000 interceptor drones monthly.
Implications: Faster, frontline-realistic testing is meant to shorten the path from prototype to fielded counter-drone capability.
Pentagon awards Casepoint up to $98.8M for AI-enabled classified legal operations
A new blanket purchase agreement makes Casepoint the exclusive provider of eDiscovery software for classified legal and data workflows across DoD.
Pentagon awards Casepoint up to $98.8M for AI-enabled classified legal operations
A new blanket purchase agreement makes Casepoint the exclusive provider of eDiscovery software for classified legal and data workflows across DoD.
- The blanket purchase agreement is worth up to $98.8 million for eDiscovery SaaS, support and training.
- Covered offices include the Office of General Counsel, DISA, and the Defense Legal Services Agency with its 28 separate offices.
- A DoD user said the platform cuts manual document-review time by 50–75%; the CEO said it does not depend on external generative-AI models.
- Casepoint claims to be the only industry provider offering IL5/IL6 authorization for eDiscovery and investigations work.
Army seeks autonomous ground vehicles to recover disabled equipment from combat zones
A new RFI calls for ruggedized autonomous ground vehicles that retrieve damaged equipment in contested areas while operating through network outages.
Army seeks autonomous ground vehicles to recover disabled equipment from combat zones
A new RFI calls for ruggedized autonomous ground vehicles that retrieve damaged equipment in contested areas while operating through network outages.
- The Army issued the request for information on June 17, 2026, seeking a "robust, ruggedized" autonomous ground vehicle for recovery missions.
- Required capabilities include autonomous operation in Denied, Degraded, Intermittent and Limited (DDIL) network conditions and rigging disabled equipment without human assistance.
- The effort cites Ukraine's demonstrated use of UGVs to recover vehicles (including other robots) with nets and hooks.
- The Army has already been experimenting with ground robots for medical evacuation and logistics resupply.
SpaceX debuts "Starfall" reusable reentry capsule on first orbital demo
SpaceX launched and deployed its disk-shaped Starfall cargo-return capsule, targeting the orbital-manufacturing retrieval market.
SpaceX debuts "Starfall" reusable reentry capsule on first orbital demo
SpaceX launched and deployed its disk-shaped Starfall cargo-return capsule, targeting the orbital-manufacturing retrieval market.
- Falcon 9 lifted off with the Starfall Demo mission from SLC-40 at Cape Canaveral at 6:53 a.m. ET on June 23, 2026.
- SpaceX confirmed deployment of the capsule in low-Earth orbit at 10:01 a.m. ET, about three hours after liftoff, with no anomalies reported.
- Starfall is a flat disk roughly 3.1 m across, 0.75 m tall, weighing ~2,100 kg, with no onboard deorbit engine.
- Aimed at returning payloads such as pharmaceuticals and orbital-manufacturing products; splashdown targeted ~700 nmi off the US West Coast.
Implications: A reusable downmass vehicle would open a recovery niche no operational US cargo capsule besides Dragon currently serves.
ESA's Euclid releases Galactic Bulge Survey, imaging 60+ million stars
ESA and the Euclid Consortium published a deep, wide-field view of the Milky Way's crowded core in their second Quick Data Release.
ESA's Euclid releases Galactic Bulge Survey, imaging 60+ million stars
ESA and the Euclid Consortium published a deep, wide-field view of the Milky Way's crowded core in their second Quick Data Release.
- The Q2 release dropped on June 24, 2026, covering the Euclid Galactic Bulge Survey across 4.8 square degrees of the inner Milky Way.
- The mosaic contains over 60 million stars, captured by Euclid's VIS camera during a ~24-hour observation on March 23, 2025.
- The dataset includes calibrated images, PSF models and photometric catalogues with ~45 million detected sources per dither down to AB magnitude 26.
- It covers the same region NASA's Roman Space Telescope will study from 2027 for microlensing planets; the image includes 51 known planetary systems.
Implications: Provides a high-resolution baseline that will sharpen Roman's planet-hunting microlensing campaign later this decade.
NASA frames Euclid bulge view as preview for Roman Space Telescope survey
NASA highlighted Euclid's new Milky Way core mosaic as a direct precursor to Roman's upcoming Galactic Bulge Time Domain Survey.
NASA frames Euclid bulge view as preview for Roman Space Telescope survey
NASA highlighted Euclid's new Milky Way core mosaic as a direct precursor to Roman's upcoming Galactic Bulge Time Domain Survey.
- NASA published its companion piece on the Euclid bulge mosaic on June 24, 2026, tied to the ESA Q2 data release.
- Roman's Galactic Bulge Time Domain Survey is planned to observe the same region over a 5-year period starting in 2027 to find microlensing exoplanets.
- The Roman telescope arrived in Florida ahead of a planned launch later in 2026 as an infrared cosmology and exoplanet observatory.
SpaceX expands Starlink with 24-satellite Falcon 9 launch from Vandenberg
A West Coast Falcon 9 added two dozen Starlink satellites to the constellation in a routine, reuse-heavy mission.
SpaceX expands Starlink with 24-satellite Falcon 9 launch from Vandenberg
A West Coast Falcon 9 added two dozen Starlink satellites to the constellation in a routine, reuse-heavy mission.
- The Starlink 17-45 mission launched from SLC-4E at Vandenberg Space Force Base, with liftoff set for 8:19 p.m. PDT on June 24, 2026.
- The flight carried 24 Starlink satellites to low-Earth orbit.
- The mission marked the 25th flight for first-stage booster B1081, which landed on a Pacific drone ship about 8.5 minutes after launch.
Qualcomm to acquire AI infrastructure startup Modular for ~$4B
Qualcomm is buying Modular to push beyond smartphones and build out its AI data-center software stack.
Qualcomm to acquire AI infrastructure startup Modular for ~$4B
Qualcomm is buying Modular to push beyond smartphones and build out its AI data-center software stack.
- Deal values Modular at roughly $4 billion, nearly 3× its September 2025 valuation of $1.6 billion.
- Modular, founded in 2022 by ex-Apple and Google engineers, makes a platform that lets AI apps run across multiple chip types without code rewrites.
- The acquisition is expected to close in the second half of 2026.
Implications: Qualcomm is signaling a serious bid to compete in AI data-center infrastructure, not just edge and mobile silicon.
.png)
Assort Health raises $120M Series C at $1.2B valuation
The healthcare voice-AI agent startup hit unicorn status as it scales patient-journey automation.
Assort Health raises $120M Series C at $1.2B valuation
The healthcare voice-AI agent startup hit unicorn status as it scales patient-journey automation.
- $120M Series C led by Menlo Ventures at a $1.2 billion valuation; total raised now exceeds $222M.
- Other investors include Lightspeed, Felicis, First Round Capital, Chemistry, Tau Ventures, Quiet Capital, and Joe Montana.
- Revenue grew 20× in the last 15 months; the platform has supported over 190M patient voice interactions to date.
Implications: Voice AI agents for healthcare front-office work (scheduling, intake, refills) are consolidating into venture-backed category leaders.
Taktile raises $110M Series C led by Goldman Sachs to automate finance decisions
The agentic decisioning platform for banks and insurers raised growth capital to expand globally.
Taktile raises $110M Series C led by Goldman Sachs to automate finance decisions
The agentic decisioning platform for banks and insurers raised growth capital to expand globally.
- $110M Series C led by Goldman Sachs Alternatives, with Tiger Global, Index Ventures, Balderton, Y Combinator and Dig Ventures; total funding now $184M.
- Founded 2020 by Maik Taro Wehmeyer and Max Eber; its modular Agentic Decision Platform combines AI agents, rules and human oversight for underwriting, claims, fraud and approvals.
- Track record cited includes 95% automation in B2B underwriting and 75% fewer AML false positives; funds target US, EMEA and LATAM expansion.
Superhuman acquires AI-detection startup GPTZero
The Grammarly-owned productivity company is buying GPTZero to build an internet "authenticity layer."
Superhuman acquires AI-detection startup GPTZero
The Grammarly-owned productivity company is buying GPTZero to build an internet "authenticity layer."
- GPTZero, founded three years ago by Edward Tian (as a Princeton senior thesis) with co-founder Alex Cui, has 19M+ registered users and $30M in ARR; terms undisclosed.
- The deal brings GPTZero's full 30-person team to Superhuman; GPTZero will continue operating as a standalone product.
- Superhuman plans to integrate GPTZero into its Superhuman Go AI assistant; Superhuman is the company formed after Grammarly acquired the email app and rebranded.
Micron beats on Q3, surges ~13% after disclosing $100B AI-memory contracts
Micron's fiscal Q3 crushed estimates on sold-out HBM demand and revealed multi-year strategic customer agreements, sending the stock sharply higher after hours.
Micron beats on Q3, surges ~13% after disclosing $100B AI-memory contracts
Micron's fiscal Q3 crushed estimates on sold-out HBM demand and revealed multi-year strategic customer agreements, sending the stock sharply higher after hours.
- Fiscal Q3 2026 (ended May 28) revenue of $41.46B, up from $9.30B a year earlier; non-GAAP EPS of $25.11 vs ~$20.49 expected.
- Shares rose about 13% after hours (to ~$1,186) after closing the regular session near $1,047; the stock is up more than 700% over the past year.
- Management disclosed transformational Strategic Customer Agreements — reported as ~$100B in minimum revenue commitments with ~$22B in customer deposits.
- Q4 guidance of about $50B revenue at ~86% gross margin; Micron's entire 2026 HBM capacity is already sold out under binding contracts.
Implications: Confirms AI-driven HBM demand is structurally repricing the memory business away from its old boom-bust cyclicality.
Cerebras tumbles ~11–20% on first post-IPO earnings as margin guidance shrinks
Cerebras posted 94% revenue growth in its debut earnings report since its May IPO, but a sharp drop in margin guidance triggered a profitability scare.
Cerebras tumbles ~11–20% on first post-IPO earnings as margin guidance shrinks
Cerebras posted 94% revenue growth in its debut earnings report since its May IPO, but a sharp drop in margin guidance triggered a profitability scare.
- Q1 2026 revenue of $193.4M, up 94% year over year.
- Full-year gross-margin guidance cut to ~38–41% (Q2 guided to ~36–38%), down from 47% in Q1 — about 1,000 bps of compression.
- Shares fell ~11% after hours on June 23 and dropped further (reported near 20%) during the June 24 session.
- CEO Andrew Feldman said investors "misunderstood" the guidance, citing temporary system rent-backs from a major customer; full-year operating margin guided to roughly -28% to -32%.
Implications: Signals investors will punish AI-hardware names that grow fast without a visible path to profitability.