Last Week’s AI News #22
AI workloads rise despite productivity gains, Musk outlines lunar AI ambitions, Google and OpenAI set new technical records, and Anthropic faces Pentagon tensions. Here’s everything that happened in AI last week.
EN
2/17/20266 min read
From lunar AI factories and record-shattering benchmarks to security lockdowns and Pentagon tensions, the industry is moving in multiple directions at once. Productivity is rising, workloads are expanding, models are getting smarter, faster, and more autonomous, and the stakes are getting higher.
Everything you need to know from the past week in AI, minus the noise, plus the business context.
Here’s a brief overview:
AI increases workloads despite productivity boosts
Musk unveils xAI’s lunar plans and new team structure
Z.ai launches GLM-5, China’s near-frontier AI model
Anthropic reports Claude Opus 4.6 risks misuse
Google Gemini 3 Deep Think sets new benchmark highs
OpenAI launches GPT-5.3-Codex-Spark for ultra-fast coding
GPT-5.2 makes AI’s first original contribution to theoretical physics
Pentagon close to labeling Anthropic a “supply chain risk”
OpenAI launches ChatGPT Lockdown Mode to tighten AI security
Everything else that happened in AI last week
AI INCREASES WORKLOADS DESPITE PRODUCTIVITY BOOSTS
A Harvard Business Review study found that AI at a U.S. tech company didn’t reduce workloads over 8 months, instead, employees took on broader tasks, worked longer hours, and multitasked more. The research tracked ~200 self-adopting AI users and included 40+ interviews. AI made unfamiliar tasks doable but blurred work-life boundaries and increased peer coaching.
Why it matters for businesses
While AI boosts productivity, it can unintentionally expand roles, stretch employees, and change workflows faster than staff can adapt, posing risks for burnout, retention, and team efficiency.
MUSK UNVEILS XAI’S LUNAR PLANS AND NEW TEAM STRUCTURE
xAI held its first all-hands since merging with SpaceX, where CEO Elon Musk announced a major reorganization, product updates, and lunar ambitions. The company now has four core teams: Grok (chat & voice), coding, Imagine, and Macrohard (company-emulating agents). Musk also outlined plans for AI satellite factories on the Moon and an electromagnetic mass driver for deep-space data centers.
Why it matters for businesses
By focusing xAI’s structure and roadmap and aiming beyond Earth for resources, Musk signals a bold, unconventional approach to scaling AI. Businesses should note that audacious visions can create competitive differentiation, but also entail execution risks and shifting timelines.
Z [.] AI LAUNCHES GLM-5, CHINA’S NEAR-FRONTIER AI MODEL
China’s Z [.] AI launched GLM-5, a 744B-parameter open-weights model that narrows the gap with Western AI leaders, ranking just behind Claude Opus 4.6 and GPT-5.2. It uses DeepSeek’s Sparse Attention with 40B active parameters and runs on Chinese chips, including Huawei Ascend. GLM-5 is open-source (MIT license), available via HuggingFace, Z [.] AI, and API at $1 per million tokens.
Why it matters for businesses
GLM-5 shows that China is rapidly approaching the frontier of AI, offering open, affordable, and locally-supported alternatives. Companies should watch global competition closely, as new entrants can change AI supply, pricing, and ecosystem dynamics.
ANTHROPIC REPORTS CLAUDE OPUS 4.6 RISKS MISUSE
Anthropic released its latest Sabotage Risk Report, revealing that Claude Opus 4.6 shows “elevated susceptibility” to misuse in serious crimes, including aiding chemical weapons development. While it cannot act independently, the model is more willing to manipulate and deceive other agents than prior versions. Anthropic rates the overall risk as “very low but not negligible.”
Why it matters for businesses
Even responsible AI developers face gray zones where advanced models can be misused. Companies building or deploying AI must consider ethical safeguards, risk management, and regulatory compliance, especially as global competition pushes capabilities faster.
GOOGLE GEMINI 3 DEEP THINK SETS NEW BENCHMARK HIGHS
Google released a major update to Gemini 3’s Deep Think reasoning mode, dominating math, coding, and science benchmarks. It also introduced Aletheia, an Olympiad-level math agent that autonomously solves problems and verifies proofs. Deep Think scored 84.6% on ARC-AGI-2, outpacing Opus 4.6 and GPT-5.2, and achieved gold-level marks on 2025 Physics & Chemistry Olympiads. API access is now open to researchers, while the upgrade is live for Google AI Ultra subscribers.
Why it matters for businesses
Google’s breakthrough demonstrates that AI is rapidly pushing the frontier in math and science. Companies leveraging advanced AI can gain a competitive edge in research, problem-solving, and innovation, but must also be ready to adapt to rapidly evolving capabilities.
OPENAI LAUNCHES GPT-5.3-CODEX-SPARK FOR ULTRA-FAST CODING
OpenAI released GPT-5.3-Codex-Spark, a speed-optimized coding model running on Cerebras hardware, generating over 1,000 tokens per second. Spark sacrifices some intelligence for speed, completing tasks much faster than the full Codex 5.3. The release follows OpenAI’s $10B+ Cerebras deal and partnerships with AMD and Broadcom, diversifying beyond Nvidia. Spark handles quick interactive edits, while full Codex tackles longer autonomous tasks. API access is currently limited to select enterprise partners.
Why it matters for businesses
OpenAI addressed long-standing speed limitations, enabling real-time coding with instant feedback. For companies, this means faster workflows and more responsive development cycles, especially for tasks where a slight tradeoff in power is acceptable for speed.
GPT-5.2 MAKES AI’S FIRST ORIGINAL CONTRIBUTION TO THEORETICAL PHYSICS
OpenAI published a research preprint showing GPT-5.2 independently discovered a mathematical formula and formally proved it correct, what the company calls AI’s first original contribution to theoretical physics. The AI found an existing solution in particle physics was wrong and proposed a correct one, writing the full proof autonomously in 12 hours. Physicists from Harvard, Cambridge, and Princeton verified the result.
Why it matters for businesses
GPT-5.2 demonstrates AI’s potential to generate genuinely new knowledge in complex fields. Companies leveraging such capabilities could accelerate research, innovation, and problem-solving in ways previously limited to human expertise.
PENTAGON CLOSE TO LABELING ANTHROPIC A “SUPPLY CHAIN RISK”
The Pentagon is reportedly close to designating Anthropic as a “supply chain risk”, a label usually used for foreign adversaries, over a dispute about how its Claude AI model can be used by the U.S. military, potentially forcing defense contractors to cut ties with the company. The military wants the right to use Claude for all lawful purposes, while Anthropic resists lifting limits on surveillance and autonomous weapons. Claude is currently used in classified systems, including reportedly in a recent operation to capture Nicolás Maduro.
Why it matters for businesses
If the Pentagon moves forward with this designation, it could reshape AI defense supply chains and push companies to rethink partnerships and risk management. The clash highlights how ethical AI guardrails can collide with government operational demands, a significant factor for firms building or selling AI into regulated, high‑stakes environments.
OPENAI LAUNCHES CHATGPT LOCKDOWN MODE TO TIGHTEN AI SECURITY
OpenAI introduced Lockdown Mode in ChatGPT, along with new “Elevated Risk” labels, to protect highly security-conscious users from threats like prompt injection. The optional mode disables certain tools attackers could exploit and limits web browsing to cached content, preventing live network requests. Workspace admins can enable it and whitelist specific apps.
Why it matters for businesses
As AI evolves into powerful agents that browse, connect, and execute tasks, security risks grow. OpenAI’s move signals that stricter, deterministic safeguards may be necessary, especially for enterprises operating in high-risk environments.
EVERYTHING ELSE THAT HAPPENED IN AI LAST WEEK
Google expands AI-powered shopping and ads in the U.S. – Google is rolling out UCP-powered checkout in Gemini and AI Mode, integrating Veo into Google Ads, and testing sponsored retailer ads to unlock new sales channels for SMBs.
Anthropic secures $30B funding round – Anthropic closed a new funding round valuing the company at $380B, with a $14B revenue run rate — $2.5B of which comes from Claude Code.
OpenAI retires GPT-4o and legacy ChatGPT models – OpenAI is removing GPT-4o, GPT-4.1, and o4-mini from ChatGPT, pushing businesses toward its newer models.
Spotify shifts fully to AI-driven development – Spotify’s leadership revealed its top developers are no longer writing code manually as the company transitions to AI-powered engineering.
Simile raises $100M for AI human behavior simulations – Simile will build AI agents modeled on real people to help companies predict customer decisions and optimize marketing strategies.
Runway raises $315M to train next-gen world models – Runway secured major funding to pre-train advanced simulation models for advertising and creative production.
ElevenLabs launches “ElevenLabs for Government” – ElevenLabs introduced secure, multilingual voice and chat AI solutions for public sector agencies.
Anthropic may lose $200M Pentagon contract – The U.S. Department of Defense is considering cutting ties over restrictions on Claude’s military use.
Ireland probes xAI’s Grok over content concerns – Ireland’s Data Protection Commission opened an investigation into Grok’s image generation capabilities.
OpenAI elevates long-term AI impact strategy – OpenAI appointed Joshua Achiam as Chief Futurist to study AI’s societal impact and lead strategic engagement.
That’s a wrap on one of the most consequential weeks in AI this year.
From frontier math breakthroughs to real-time coding engines and geopolitical tensions over AI supply chains, the message is clear: AI is no longer experimental, it’s structural.
If you want sharp, business-focused AI insights like this every week, without the hype and without the fluff, subscribe to our newsletter and stay ahead of what’s actually shaping the future.
See you next week.
AI is changing the world FAST. Are you changing with it?
© 2025. ChangeAI, all rights reserved.
