AI News: August 4, 2026

1. Alibaba Releases Qwen3.8-Max, a 2.4-Trillion-Parameter Flagship

Alibaba. Alibaba unveiled Qwen3.8-Max, its largest model yet, a mixture-of-experts system with 2.4 trillion total parameters (95 billion active) and a 1 million token context window, available through Alibaba Cloud Model Studio APIs with open weights due the following week. Early benchmarks place it around fifth in Text Arena and second in Vision Arena while pricing well below Claude Opus 5. The release pushes open-weight models further into the long-horizon agentic territory that closed frontier labs have dominated. Source

2. MiniMax H3 Becomes the First Open Model to Top an AI Video Ranking

MiniMax. China’s MiniMax H3 became the first open-weight model to reach the top of an AI video generation leaderboard, a category long led by proprietary systems. The result signals that open video generators are closing the quality gap with closed competitors. It gives teams building video pipelines a credible option they can self-host or fine-tune. Source

3. Palantir Q2 Revenue Jumps 93% on AI Demand, Guidance Raised

Palantir. Palantir reported second-quarter 2026 revenue of about $1.94 billion, up 93 percent year over year, with US commercial revenue up 149 percent, and raised full-year guidance to roughly $8.15 billion. Shares climbed in after-hours trading as the company credited surging enterprise demand for its AI platform. The numbers are a closely watched proxy for how much revenue enterprise AI deployments are actually generating. Source

4. Two Teams Solve the Same Quantum Crypto Problem With GPT-5.6, Hours Apart

Research. Two research groups independently solved the same quantum cryptography problem using OpenAI’s GPT-5.6 just three hours apart, an unusually direct example of AI-accelerated parallel discovery. The near-simultaneous results raise questions about credit and priority when frontier models compress research timelines. For practitioners, it hints that model-assisted proofs may increasingly collide on the same open problems. Source

5. IBM: 92% of AI-Breached Firms Lacked Basic Access Controls

IBM. An IBM security report found that 92 percent of companies that suffered AI-related breaches had no basic access controls on their AI systems. The finding points to a widening governance gap as organizations rush models into production without foundational identity and permission safeguards. It reframes AI security as an access-management problem first, not just a model-safety one. Source

6. Interpol Calls AI the Core Operational Driver of Cybercrime in Africa

Interpol. Interpol said AI has become the core operational driver of cybercrime across Africa, citing a rise in AI-enabled scams, phishing, and fraud automation. The assessment shifts from treating AI as an emerging threat to naming it as central to how criminal operations now run. It adds pressure on defenders to match AI-scaled attacks with AI-aware detection. Source

7. A Real $200K macOS Flaw Went Unreported Amid Bug-Bounty AI Slop

Apple. A genuine macOS vulnerability estimated to be worth $200,000 reportedly went unreported because Apple’s bug bounty inbox was flooded with low-quality AI-generated submissions. The case shows how AI-generated noise can overwhelm security triage and bury real findings. It is an early warning that high-volume AI slop can degrade the signal in vulnerability disclosure programs. Source

8. AI-Proctored Remote Exam Fails, 58,000 Students Must Retake

Education. An AI-proctored remote exam malfunctioned badly enough that authorities ordered roughly 58,000 students to retake it. The failure is a concrete cautionary case on the reliability of AI supervision in high-stakes settings. It underscores that automated proctoring can create large-scale operational and fairness risks when it misfires. Source

9. Auterion to Give 50,000 Ukrainian Drones AI Targeting in $100M Deal

Auterion. A $100 million deal will equip roughly 50,000 Ukrainian Shrike FPV drones with US firm Auterion’s Skynode S autonomy kits, enabling fire-and-forget terminal guidance that tracks moving targets without an operator. Deliveries began in mid-July and continue over the coming months. The scale marks one of the largest fielded deployments of on-board AI targeting to date. Source

10. Apple Ships Its Revamped Siri to a Muted Reception

Apple. Apple shipped its long-delayed Siri overhaul, but early coverage describes the launch as anticlimactic relative to years of expectations. Reviewers note the revamped assistant arrives after rivals have already normalized capable voice and agent features. For Apple, the muted response raises the bar for proving the new Siri is more than catch-up. Source

11. AWS Backs Vibe-Coding Startup Superblocks

AWS. Amazon Web Services is backing vibe-coding startup Superblocks, a tie-up that TechCrunch frames as strategically significant for AWS’s position in AI-assisted software development. The move signals AWS leaning into natural-language app building rather than ceding it to standalone tools. It is a bet that AI coding demand will pull through underlying cloud consumption. Source

12. ChatGPT Is Congress’s Most-Used AI Tool

Government. New data shows ChatGPT is the most widely used AI tool among members of the US Congress and their staff, outpacing rival assistants on Capitol Hill. The adoption underscores how default consumer AI tools are seeping into government workflows ahead of formal procurement or policy. It also raises questions about data handling and oversight for official use. Source