Responsible AI · Enterprise engineering · Global delivery
Curated analysis of AI, ML, and Agentic developments.
Apr 23, 2026
Released 23 April 2026, GPT-5.5 is OpenAI's first new base model in a year, with a 400K context in Codex, strong agentic benchmarks, a high cyber and bio risk rating and tighter controls on sensitive requests.
Apr 14, 2026
OWASP's GenAI Security Project published its Q1 2026 exploit round-up on 14 April, cataloguing eight incidents from a 150 GB government breach to agent-driven data loss, and finding most stemmed from design flaws rather than CVEs.
Apr 13, 2026
The 2026 AI Index, published 13 April, records global AI investment above $581 billion in 2025, near-total industry dominance of frontier model development, rising compute and emissions, and modest public trust in regulators.
Apr 07, 2026
On 7 April 2026 Anthropic unveiled Claude Mythos Preview, a model it says has found thousands of zero-day vulnerabilities, and restricted access to a defensive-security coalition backed by $100 million in credits.
Mar 31, 2026
On 31 March 2026 a source map in Claude Code v2.1.88 exposed roughly 512,000 lines of Anthropic's agent code; a concurrent supply-chain attack on the axios package pushed a remote access trojan to unlucky updaters.
Mar 20, 2026
On 20 March 2026 the White House published a National Policy Framework for AI that urges Congress to override state AI regulation, favours light-touch federal rules and adds child-safety and deepfake measures.
Mar 16, 2026
At GTC on 16 March 2026 Nvidia announced that all seven chips of its Vera Rubin platform are in full production, promising large gains in inference per watt and shipping through cloud and OEM partners in the second half of 2026.
Mar 13, 2026
On 13 March 2026 the Council of the EU adopted its negotiating position on the Digital Omnibus on AI, proposing to delay high-risk obligations to December 2027 and August 2028 and adding a new prohibited practice.
Mar 05, 2026
OpenAI released GPT-5.4 on 5 March 2026 in standard, Thinking and Pro variants, adding native computer use, a 1M-token context and a Tool Search feature, while keeping a high cyber-risk classification.
