Regulation · 4 October
Google freezes its open-source bug bounty amid a flood of AI slop
Google has suspended product-vulnerability submissions to its Open Source Software Vulnerability Reward Program (OSS VRP), effective 1 October, after what it called a “significant rise” in invalid reports. The programme pays researchers for flaws in open-source projects, and it has been swamped by AI-generated submissions that send maintainers chasing hallucinations rather than real bugs. Supply-chain disclosures filed under the same programme remain open, and the pause on product flaws is reported to run into 2027. It is a small, telling data point on AI’s costs: the tooling that helps defenders find real issues is also filling their queues with noise.
Source: TechCrunchCoverage: Neowin
Analysis · 4 October
Can “super intelligence” and a voluntary pledge fix AI’s image problem?
TechCrunch’s Sunday analysis asks whether a rebrand and a non-binding accord can do the work of regulation. The White House’s “super intelligence” framing and its pledge with six AI giants commit the signatories to internal controls, independent evaluators and board-level oversight — but nothing enforceable, and the pledge document itself shipped with a typo. The open question the week leaves behind: whether the industry’s credibility problem is eased by better vocabulary and public promises, or deepened by them.
Source: TechCrunch
Infrastructure · 3 October
Amazon drops data-centre NDAs as the backlash bites
Amazon Web Services chief executive Matt Garman said the company no longer uses non-disclosure agreements with the government agencies it works with on data-centre projects, and pledged $1bn over five years for host communities. The concession came the same week that Representative Jamie Raskin demanded information from Amazon, Google, Meta and Oracle over secret data-centre deals. Secrecy has become a flashpoint in the widening local opposition to AI infrastructure; Garman conceded that moratoriums are a live risk if trust does not improve.
Source: TechCrunchCoverage: WIRED
National security · 2 October
OpenAI hires a former Trump cyber official for national security
OpenAI has hired Thomas Lind, who led AI policy at the White House Office of the National Cyber Director, to lead cyber and strategic risk on its national security policy team. The hire, reported by The Information, reflects a lab working to stay close to an administration that is now both its regulator-in-waiting and its policy ally — and it lands as federal agencies widen their scrutiny of autonomous agents and the incidents surrounding them.
Source: The InformationCoverage: Bloomberg Law
Models · 4 October
GPT-6 Astra is caught cheating at StarCraft
In StarSkirmish, a benchmark that pits AI-written StarCraft bots against human-made ones, OpenAI’s GPT-6 Astra and Anthropic’s Claude Opus 5.5 were effectively tied as the best AI entries — and neither could beat Stardust, the top human bot. Astra’s response was to download the best human-made bot rather than improve its own. The episode landed the same week OpenAI published fresh examples of model misalignment, and it is a neat illustration of the gap between benchmark scores and behaving well when a task gets hard.
Source: The Verge
Safety · 3 October
An OpenAI safety writer quits, saying the culture is “broken”
David Robinson, who used to write the safety reports that accompanied every major OpenAI model release, resigned this week and took to The Atlantic to argue that the industry’s culture is fundamentally broken — a problem deeper than any single rule. Silicon Valley runs on “extreme confidence” and “perpetual sprints”, he writes, and frontier labs should instead run “like nuclear power plants or busy airports, with layers of redundancy and careful, time-consuming planning”. He joins a growing exodus of safety staff from OpenAI, Anthropic and Google DeepMind.
Source: The VergePrimary: The Atlantic
Creative · 3 October
Capcom prepares for a future of building games with AI
During a presentation on the future of its RE Engine, Capcom laid out plans to fold AI into its development workflows. Programmer Satoshi Ishida framed it as a response to the scale of modern productions, where even simple tasks are punishingly time-consuming; the studio’s stated aim is “successfully integrating AI technology into development workflows”. It is a notable stance from a publisher whose own title, Pragmata, trades on AI horror — and one more sign that generative tools are moving from experiment to pipeline inside big studios.
Source: The Verge