Pentagon demands Anthropic remove Claude AI safeguards in military dispute - The Pentagon is escalating a dispute with Anthropic over how the military can use Claude, including a reported deadline and threats that could range from cutting ties to declaring the company a “supply chain risk.” Reporting also describes discussion of a Defense Production Act approach that could compel changes to how Claude is provided for defense needs.

Why it matters: The outcome may decide whether AI guardrails are real limits or negotiable features. If governments can override safety restrictions once AI becomes operational infrastructure, future safeguards may be shaped less by engineers and researchers — and more by procurement power, national security priorities, and competitive pressure among AI vendors.

I hacked ChatGPT and Google's AI - and it only took 20 minutes - A BBC investigation found that widely used AI systems, including ChatGPT and Google’s AI tools, could be manipulated in under 20 minutes using relatively simple prompt-engineering techniques. The experiment demonstrated how safeguards can sometimes be bypassed, raising questions about how resilient current AI protections are against misuse or adversarial inputs.

Why it matters: As AI becomes embedded in everyday workflows and decision-making, security and robustness are becoming just as important as model capability — highlighting the growing need for stronger safety testing and real-world stress evaluations.

Meta and AMD Partner for Longterm AI Infrastructure Agreement - Meta and AMD announced a long-term agreement to collaborate on AI infrastructure, with AMD providing advanced chips and hardware systems to support Meta’s expanding AI workloads. The partnership reflects Meta’s continued effort to diversify beyond Nvidia-dominated supply chains while scaling compute capacity for large-scale AI models and services.

Why it matters: AI development is increasingly shaped by infrastructure alliances, not just model innovation — and diversified chip ecosystems could reshape competition, pricing, and access to AI compute across the industry.

Why SWE-bench Verified no longer measures frontier coding capabilities - OpenAI announced it will no longer rely on the SWE-Bench Verified benchmark to evaluate frontier coding models, arguing that the test no longer accurately measures modern AI coding capabilities. As models improve, older benchmarks can become saturated or fail to capture real-world performance, prompting the shift toward newer evaluation methods.

Why it matters: The change highlights a broader industry challenge: AI capabilities are advancing faster than the benchmarks used to measure them, making reliable evaluation — and meaningful model comparison — increasingly difficult.

Find a company website you want to analyze. Copy the homepage URL (or multiple pages if available).

Open ChatGPT and paste the link, then prompt:
“Analyze this website as a business consultant. Identify the target audience, core value proposition, messaging strengths, and weaknesses.”

Next, ask ChatGPT to evaluate positioning:
“Based on this website, what problem does this company solve, and how clearly is that communicated?”

To generate actionable insights, prompt:
“Suggest three improvements that could increase conversions or clarity without changing the company’s core offering.”

If you want deeper analysis, ask:
“Compare this website’s messaging to industry best practices and identify gaps.”

Now you have a rapid strategic audit you can use for research, competitive analysis, or content ideas.

Descript: Descript is an all-in-one audio and video editing tool that lets you edit media as easily as a text document, with AI features for transcription, overdubbing, and streamlined content production. We’re excited to invite you to join Descript’s Creator Plan and enjoy 50% off your first 2 months. Be sure to use our link to activate the discount! (As an affiliate partner with Descript, AiNews may earn a small commission if you sign up using our link. There’s no extra cost to you — it simply helps support our independent reporting and the work we do.)

HeyVid: Transform any text to video with AI.

PropLab: Know Your Exact Offer in 60 Seconds — Before You Call the Agent. PropLab is AI-powered software to calculate ARV, estimate rehab costs, & underwrite real estate deals — so you never make blind offers.

SecureSaas: Website Vulnerability Scanner for Web Apps. Run a free website vulnerability scan on your web app in seconds. Our automated scanner crawls your site, checks for security issues, and delivers a detailed report with severity scores — upgrade for actionable fixes.