Anthropic Ban Lifts, Sonnet, Science Launch.

Mazen Al Sabbagh

— 5 min read
Share

TL;DR

  • Export Controls Lifted: The Commerce Department rescinded its June 12 export control on Claude Fable 5 and Mythos 5, with Anthropic commencing global access restoration this Wednesday.
  • Anthropic's Dual Launch: Coincident with the lifting of sanctions, Anthropic introduced Claude Sonnet 5, offering near-Opus capabilities, and Claude Science, a dedicated AI research workbench.
  • Gemini 3.5 Pro Delayed: Google's Gemini 3.5 Pro officially missed its June general availability target, a slip predicted with 97% certainty on Polymarket.
  • Youth Employment Impact: The Stanford/ADP Canaries Dashboard reveals a 3.8% annual contraction in AI-exposed job sectors for the 22-25 age demographic.
  • Half-Year Market Close: Equities rebounded into the close of H1, with the S&P 500 up 9.6% and the Nasdaq gaining 12%, propelled by an 80%+ surge in semiconductors.

Lead Story: Anthropic Ban Lifts, Sonnet, Science Launch.

Eighteen days post-order from Commerce Secretary Howard Lutnick to pull Claude Fable 5 and Mythos 5 from international markets, the Commerce Department has lifted the export controls on both models. Anthropic confirmed this on X, indicating access restoration begins Wednesday, with a subsequent comprehensive update. The June 12 directive, representing the first emergency export control applied to a deployed AI model, was initiated following a reported jailbreak that exposed Mythos-class cyber capabilities beneath Fable 5's established guardrails.

Signs of this thaw were visible for days. Washington had already allowed Mythos 5 back for select trusted US organizations on June 26, and Forbes reported Monday on an imminent return, with an administration source informing Axios of Anthropic's "positive collaboration with the government." The shift in negotiation leadership from Dario Amodei to co-founder and chief compute officer Tom Brown—a move from safety-brand diplomacy to technical resolution—appears to have been effective. Anticipate restored models to feature reinforced guardrails and enhanced identity verification.

Anthropic did not defer its agenda. The company also launched Claude Sonnet 5 today, positioned as its most agentic Sonnet iteration. This model demonstrates planning capabilities, drives browser and terminal interactions, and achieves a 63.2% score on agentic coding, compared to Opus 4.8's 69.2% and Sonnet 4.6's 58.1%. Intriguingly, it outperforms Opus 4.8 on specific knowledge-work tasks. Introductory pricing is set at $2 per million input tokens and $10 output through August 31 (thereafter $3/$15), becoming the default for Free and Pro users this week. Concurrently, Anthropic unveiled Claude Science, an AI research workbench integrated with over 60 scientific databases, alongside its initiation of proprietary drug-discovery programs for neglected diseases. A company that spent most of June as a cautionary tale on regulatory power concluded the month by defining the day's entire narrative.

In Other News

Gemini 3.5 Pro's Missed Deadline. Google's flagship model failed to meet its June general availability commitment, a delay that Polymarket had priced at 97% probability. Google attributes this to extensive token consumption in agentic tasks, pushing the launch to July, with a 2-million-token context window and Deep Think exclusive to the $250/month Ultra tier. This setback follows a significant talent drain, notably Noam Shazeer to OpenAI and key researchers to Anthropic. Google did release new image-generation models and free Nano Banana personalization for US Gemini users today.

Quantifying AI's Impact on Early Career Employment. The Canaries Dashboard, a new Stanford and ADP initiative, quantifies the erosion of entry-level roles. Data from 4.6 million workers shows employment for the 22-25 age bracket in AI-exposed sectors declining 3.8% annually, contrasting with a 2% growth in least-exposed jobs. While the aggregate effect across all demographics is a modest -0.2%, the acute impact on early career professionals is evident. Erik Brynjolfsson, an early proponent of this view, notes this decline persists without reversal.

H1 Market Close: A Rebound After Volatility. Equities recovered significantly into the June 30 close, following the Nasdaq's recent five-session downturn. Nvidia, AMD, and Intel saw gains of 2.6%, 7.7%, and 6% respectively. The first-half figures are robust: S&P 500 +9.6%, Nasdaq +12%, Dow +8.9%, and the Russell 2000 leading with +22%—its strongest H1 since 1991. Semiconductor stocks surged over 80%. Deutsche Bank attributed the late-June volatility among the Magnificent Seven to factors including positioning adjustments, hyperscaler capital expenditure concerns, a hawkish Federal Reserve, and increasing chip input costs.

Copilot's Billing Shift Signals End of Flat-Rate AI Coding. June concluded with significant repercussions from GitHub Copilot's shift to usage-based billing. Developers reported bill increases ranging from 10x to 50x, with individual agentic sessions consuming $30-40 in credits. CPO Mario Rodriguez acknowledged that "Copilot is not the same product it was a year ago." This move, following Cursor and Windsurf, effectively terminates the flat-rate business model for agentic coding tools.

X / Social Pulse

Anthropic's succinct X post—"We've received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5"—immediately captured the AI discourse. Simon Willison's Sonnet 5 analysis provided the day's most incisive observation: the new tokenizer generates approximately 30% more tokens for equivalent English text, effectively an unannounced price hike. He also noted the model misidentified a pelican as a goose. Concurrently, GitHub's new billing announcement garnered over 900 downvotes, while Polymarket participants celebrated their accurate prediction regarding the Gemini delay.

One to Watch

OpenAI's GPT-5.6 General Release. OpenAI's Sol, Terra, and Luna tiers remain exclusive to approximately 20 government-vetted entities since their June 26 preview. Sam Altman has indicated broad availability within "a couple of weeks." A July 2 interim milestone under the June 2 executive order could expedite this timeline. With Sol achieving a 91.9% terminal-bench score against Mythos 5's 88.0%, its full release promises to re-ignite the frontier AI performance benchmarks.

Quick Hits

  • China PMI Rises: Factory PMI in China increased to 50.3 in June, driven by robust AI-related chip and computer export demand.
  • Geofence Warrants Ruled Searches: The Supreme Court determined that broad geofence warrants constitute Fourth Amendment searches, limiting bulk location-data acquisition.
  • Indonesia AI Data Center: Firmus and Nvidia finalized an 8-year agreement for a 360-megawatt AI data center in Batam, Indonesia, projected to house up to 170,000 GPUs by 2028.
  • DRAM Price Surge Forecast: Jefferies predicts DRAM price increases of 40–50% in Q3 and 30–40% in Q4, with AI server demand consuming 60–70% of supply, signaling no market relief until 2028.
  • Meta's Chatbot Probes: Wired details Meta's engagement of hundreds of contractors, posing as minors, to test Google and OpenAI chatbots for vulnerabilities on sensitive subjects.

The swift resolution of the export control saga marks a significant, albeit peculiar, precedent in the government-frontier AI lab dynamic. Washington demonstrated its capacity to de-platform a deployed model with immediate effect, while Anthropic showcased its ability to navigate complex regulatory negotiations while simultaneously advancing its product roadmap. As July commences, the competitive landscape is fully reset, with every key player's flagship model either reinstated, newly introduced, or on the cusp of release.

Sources

Anthropic: CNBC — export controls lifted · Anthropic on X · Forbes — return imminent · Forbes — June 12 order background · Anthropic — Sonnet 5 · TechCrunch — Sonnet 5 · Anthropic — Claude Science · STAT — Claude Science · TechCrunch — Claude Science

Google: Build Fast with AI — June 30 roundup · unrot — June 30 top 10

Markets: CNBC — June 30 session · CNBC — late-June selloff

Labor: Fortune — Brynjolfsson interview · Stanford Digital Economy Lab — Canaries Dashboard

Other: Simon Willison — Sonnet 5 notes · Tech Startups — June 30 roundup

Lock in. M. mazen@thorterminal.com

Read more